https://vulnerability.circl.lu Contains only the most 10 recent sightings. http://www.rssboard.org/rss-specification python-feedgen en Wed, 02 Sep 2026 08:33:52 +0000 https://vulnerability.circl.lu/sighting/30078121-f4a9-41bc-98b8-26f6441e6235/export {"uuid": "30078121-f4a9-41bc-98b8-26f6441e6235", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "c933734a-9be8-4142-889e-26e95c752803", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://vulnerability.circl.lu/bundle/d1d68b37-fe59-4031-ba0b-bd808b0975dc", "content": "", "creation_timestamp": "2026-08-20T06:40:22.599637Z"} {"uuid": "30078121-f4a9-41bc-98b8-26f6441e6235", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "c933734a-9be8-4142-889e-26e95c752803", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://vulnerability.circl.lu/bundle/d1d68b37-fe59-4031-ba0b-bd808b0975dc", "content": "", "creation_timestamp": "2026-08-20T06:40:22.599637Z"} https://vulnerability.circl.lu/sighting/30078121-f4a9-41bc-98b8-26f6441e6235/export Thu, 20 Aug 2026 06:40:22 +0000 https://vulnerability.circl.lu/sighting/a3d84873-80c5-4398-985b-b35177ba1be8/export {"uuid": "a3d84873-80c5-4398-985b-b35177ba1be8", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "exploited", "source": "https://t.me/bhhub/1211", "content": "Weekly 8 AI & Cyber signals to act on (Aug 10\u201317, 2026)\n\n#AISecurity@bhhub\n\n\u2728 Encrypted reasoning traces leak across LLM sessions\nClient-returned AI reasoning is a cross-user attack surface: researchers moved encrypted blocks into a weaker model from the same provider and recovered plaintext across Anthropic, OpenAI, and Google. Decoding 315,320 public-repository blocks exposed 367 PII artifacts and 182 credentials, plus hidden hazardous reasoning. Bind traces to user, session, and model; this is a preprint.\n\n\u2728 PIPES gives agent observations provenance\nAI agents can mistake attacker-controlled tool fields for authoritative state, making unsafe actions look justified. PIPES checks response units against semantic priors and provenance; on six splits it cut Gemma 4 31B attack success from 84.7% to 2.3% while benign utility rose from 90.6% to 92.5%. The preprint tests two agents and assumes trusted metadata.\n\n\u2728 ColluSkill hides one attack across three skills\nLLM planning turns plausible agent skills into one malicious workflow: ColluSkill splits intent across packages, then refines them against scanners. On 200 three-skill chains and six scanners, attack success reached 96.0%; ChainGuard reduced it to 22.5% while passing 99.5% of benign workflows. Scan capability and artifact flow across the installed set.\n\n#AppSec@bhhub\n\n\u2728 PySeType finds nine confirmed zero-days\nLLMs supply semantic types that static analysis cannot infer from syntax: PySeType asks models to check natural-language security meanings, then traverses Python control flow with pytype. Across 103 web projects it reports 87% precision, 88% accuracy, and 15 potential zero-days, nine developer-confirmed. Dynamic loading, callbacks, and cross-language flows remain blind spots.\n\n\u2728 SRE-Bench tests agents on unseen binaries\nAI security capability drops when source code becomes a protected binary: SRE-Bench contributes 19 private programs, 44 in-house anti-analysis primitives, 262 instances, and 1,572 deterministic tasks built over 5,000 expert hours. The best of five frontier models scored 61.4% per instance but fully solved only 31.5%. Binary reverse engineering needs contamination control, realistic scale, and protection-aware evals.\n\n#RedTeam@bhhub\n\n\u2728 ZOOMSDAY: an AI agent finds a zero-click RCE chain\nAn AI coding agent materially accelerated closed-source exploit research: A Security reports finding and exploiting Zoom\u2019s annotation parser in under 24 hours and fewer than 20 prompts. A crafted 745-byte packet reaches a 128-byte stack buffer via an unchecked count; Zoom corroborates CVE-2026-53413/414/415 and patched affected clients. The RCE demo is macOS and the AI-speed claim is first-party, but shared code makes patching urgent.\n\n\u2728 MazeRunner branches through black-box pentests\nThree LLM agents preserve clues, revise failed actions, and switch attack branches instead of following one linear pentest plan. On ten recent HTB targets under a 20M-token cap, MazeRunner completed 47.7% of subtasks versus 36.2% for PentestGPT-V2, reached user access on six targets, and root on two; both baselines reached user on two and root on none. Human approval is still needed for destructive live actions.\n\n#BlueTeam@bhhub\n\n\u2728 GraphRAG makes threat hunts survive IOC rotation\nKnowledge-graph retrieval pushes AI-generated hunt plans beyond disposable indicators toward behaviors and techniques. In an APT28 case, GraphRAG preserved 100% of detections after every IP, domain, and hash was rotated; naive RAG preserved 29%. Nine reports from four vendors showed the same direction, but prompt wording mattered nearly as much and the corpus is small. Evaluate detection durability, not rule count.", "creation_timestamp": "2026-08-18T00:00:29.221584Z"} {"uuid": "a3d84873-80c5-4398-985b-b35177ba1be8", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "exploited", "source": "https://t.me/bhhub/1211", "content": "Weekly 8 AI & Cyber signals to act on (Aug 10\u201317, 2026)\n\n#AISecurity@bhhub\n\n\u2728 Encrypted reasoning traces leak across LLM sessions\nClient-returned AI reasoning is a cross-user attack surface: researchers moved encrypted blocks into a weaker model from the same provider and recovered plaintext across Anthropic, OpenAI, and Google. Decoding 315,320 public-repository blocks exposed 367 PII artifacts and 182 credentials, plus hidden hazardous reasoning. Bind traces to user, session, and model; this is a preprint.\n\n\u2728 PIPES gives agent observations provenance\nAI agents can mistake attacker-controlled tool fields for authoritative state, making unsafe actions look justified. PIPES checks response units against semantic priors and provenance; on six splits it cut Gemma 4 31B attack success from 84.7% to 2.3% while benign utility rose from 90.6% to 92.5%. The preprint tests two agents and assumes trusted metadata.\n\n\u2728 ColluSkill hides one attack across three skills\nLLM planning turns plausible agent skills into one malicious workflow: ColluSkill splits intent across packages, then refines them against scanners. On 200 three-skill chains and six scanners, attack success reached 96.0%; ChainGuard reduced it to 22.5% while passing 99.5% of benign workflows. Scan capability and artifact flow across the installed set.\n\n#AppSec@bhhub\n\n\u2728 PySeType finds nine confirmed zero-days\nLLMs supply semantic types that static analysis cannot infer from syntax: PySeType asks models to check natural-language security meanings, then traverses Python control flow with pytype. Across 103 web projects it reports 87% precision, 88% accuracy, and 15 potential zero-days, nine developer-confirmed. Dynamic loading, callbacks, and cross-language flows remain blind spots.\n\n\u2728 SRE-Bench tests agents on unseen binaries\nAI security capability drops when source code becomes a protected binary: SRE-Bench contributes 19 private programs, 44 in-house anti-analysis primitives, 262 instances, and 1,572 deterministic tasks built over 5,000 expert hours. The best of five frontier models scored 61.4% per instance but fully solved only 31.5%. Binary reverse engineering needs contamination control, realistic scale, and protection-aware evals.\n\n#RedTeam@bhhub\n\n\u2728 ZOOMSDAY: an AI agent finds a zero-click RCE chain\nAn AI coding agent materially accelerated closed-source exploit research: A Security reports finding and exploiting Zoom\u2019s annotation parser in under 24 hours and fewer than 20 prompts. A crafted 745-byte packet reaches a 128-byte stack buffer via an unchecked count; Zoom corroborates CVE-2026-53413/414/415 and patched affected clients. The RCE demo is macOS and the AI-speed claim is first-party, but shared code makes patching urgent.\n\n\u2728 MazeRunner branches through black-box pentests\nThree LLM agents preserve clues, revise failed actions, and switch attack branches instead of following one linear pentest plan. On ten recent HTB targets under a 20M-token cap, MazeRunner completed 47.7% of subtasks versus 36.2% for PentestGPT-V2, reached user access on six targets, and root on two; both baselines reached user on two and root on none. Human approval is still needed for destructive live actions.\n\n#BlueTeam@bhhub\n\n\u2728 GraphRAG makes threat hunts survive IOC rotation\nKnowledge-graph retrieval pushes AI-generated hunt plans beyond disposable indicators toward behaviors and techniques. In an APT28 case, GraphRAG preserved 100% of detections after every IP, domain, and hash was rotated; naive RAG preserved 29%. Nine reports from four vendors showed the same direction, but prompt wording mattered nearly as much and the corpus is small. Evaluate detection durability, not rule count.", "creation_timestamp": "2026-08-18T00:00:29.221584Z"} https://vulnerability.circl.lu/sighting/a3d84873-80c5-4398-985b-b35177ba1be8/export Tue, 18 Aug 2026 00:00:29 +0000 https://vulnerability.circl.lu/sighting/dda4a57a-f7d3-4aac-a45b-61be764e2a75/export {"uuid": "dda4a57a-f7d3-4aac-a45b-61be764e2a75", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://t.me/bhhub/1211", "content": "Weekly 8 AI & Cyber signals to act on (Aug 10\u201317, 2026)\n\n#AISecurity@bhhub\n\n\u2728 Encrypted reasoning traces leak across LLM sessions\nClient-returned AI reasoning is a cross-user attack surface: researchers moved encrypted blocks into a weaker model from the same provider and recovered plaintext across Anthropic, OpenAI, and Google. Decoding 315,320 public-repository blocks exposed 367 PII artifacts and 182 credentials, plus hidden hazardous reasoning. Bind traces to user, session, and model; this is a preprint.\n\n\u2728 PIPES gives agent observations provenance\nAI agents can mistake attacker-controlled tool fields for authoritative state, making unsafe actions look justified. PIPES checks response units against semantic priors and provenance; on six splits it cut Gemma 4 31B attack success from 84.7% to 2.3% while benign utility rose from 90.6% to 92.5%. The preprint tests two agents and assumes trusted metadata.\n\n\u2728 ColluSkill hides one attack across three skills\nLLM planning turns plausible agent skills into one malicious workflow: ColluSkill splits intent across packages, then refines them against scanners. On 200 three-skill chains and six scanners, attack success reached 96.0%; ChainGuard reduced it to 22.5% while passing 99.5% of benign workflows. Scan capability and artifact flow across the installed set.\n\n#AppSec@bhhub\n\n\u2728 PySeType finds nine confirmed zero-days\nLLMs supply semantic types that static analysis cannot infer from syntax: PySeType asks models to check natural-language security meanings, then traverses Python control flow with pytype. Across 103 web projects it reports 87% precision, 88% accuracy, and 15 potential zero-days, nine developer-confirmed. Dynamic loading, callbacks, and cross-language flows remain blind spots.\n\n\u2728 SRE-Bench tests agents on unseen binaries\nAI security capability drops when source code becomes a protected binary: SRE-Bench contributes 19 private programs, 44 in-house anti-analysis primitives, 262 instances, and 1,572 deterministic tasks built over 5,000 expert hours. The best of five frontier models scored 61.4% per instance but fully solved only 31.5%. Binary reverse engineering needs contamination control, realistic scale, and protection-aware evals.\n\n#RedTeam@bhhub\n\n\u2728 ZOOMSDAY: an AI agent finds a zero-click RCE chain\nAn AI coding agent materially accelerated closed-source exploit research: A Security reports finding and exploiting Zoom\u2019s annotation parser in under 24 hours and fewer than 20 prompts. A crafted 745-byte packet reaches a 128-byte stack buffer via an unchecked count; Zoom corroborates CVE-2026-53413/414/415 and patched affected clients. The RCE demo is macOS and the AI-speed claim is first-party, but shared code makes patching urgent.\n\n\u2728 MazeRunner branches through black-box pentests\nThree LLM agents preserve clues, revise failed actions, and switch attack branches instead of following one linear pentest plan. On ten recent HTB targets under a 20M-token cap, MazeRunner completed 47.7% of subtasks versus 36.2% for PentestGPT-V2, reached user access on six targets, and root on two; both baselines reached user on two and root on none. Human approval is still needed for destructive live actions.\n\n#BlueTeam@bhhub\n\n\u2728 GraphRAG makes threat hunts survive IOC rotation\nKnowledge-graph retrieval pushes AI-generated hunt plans beyond disposable indicators toward behaviors and techniques. In an APT28 case, GraphRAG preserved 100% of detections after every IP, domain, and hash was rotated; naive RAG preserved 29%. Nine reports from four vendors showed the same direction, but prompt wording mattered nearly as much and the corpus is small. Evaluate detection durability, not rule count.", "creation_timestamp": "2026-08-17T08:00:03.938053Z"} {"uuid": "dda4a57a-f7d3-4aac-a45b-61be764e2a75", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://t.me/bhhub/1211", "content": "Weekly 8 AI & Cyber signals to act on (Aug 10\u201317, 2026)\n\n#AISecurity@bhhub\n\n\u2728 Encrypted reasoning traces leak across LLM sessions\nClient-returned AI reasoning is a cross-user attack surface: researchers moved encrypted blocks into a weaker model from the same provider and recovered plaintext across Anthropic, OpenAI, and Google. Decoding 315,320 public-repository blocks exposed 367 PII artifacts and 182 credentials, plus hidden hazardous reasoning. Bind traces to user, session, and model; this is a preprint.\n\n\u2728 PIPES gives agent observations provenance\nAI agents can mistake attacker-controlled tool fields for authoritative state, making unsafe actions look justified. PIPES checks response units against semantic priors and provenance; on six splits it cut Gemma 4 31B attack success from 84.7% to 2.3% while benign utility rose from 90.6% to 92.5%. The preprint tests two agents and assumes trusted metadata.\n\n\u2728 ColluSkill hides one attack across three skills\nLLM planning turns plausible agent skills into one malicious workflow: ColluSkill splits intent across packages, then refines them against scanners. On 200 three-skill chains and six scanners, attack success reached 96.0%; ChainGuard reduced it to 22.5% while passing 99.5% of benign workflows. Scan capability and artifact flow across the installed set.\n\n#AppSec@bhhub\n\n\u2728 PySeType finds nine confirmed zero-days\nLLMs supply semantic types that static analysis cannot infer from syntax: PySeType asks models to check natural-language security meanings, then traverses Python control flow with pytype. Across 103 web projects it reports 87% precision, 88% accuracy, and 15 potential zero-days, nine developer-confirmed. Dynamic loading, callbacks, and cross-language flows remain blind spots.\n\n\u2728 SRE-Bench tests agents on unseen binaries\nAI security capability drops when source code becomes a protected binary: SRE-Bench contributes 19 private programs, 44 in-house anti-analysis primitives, 262 instances, and 1,572 deterministic tasks built over 5,000 expert hours. The best of five frontier models scored 61.4% per instance but fully solved only 31.5%. Binary reverse engineering needs contamination control, realistic scale, and protection-aware evals.\n\n#RedTeam@bhhub\n\n\u2728 ZOOMSDAY: an AI agent finds a zero-click RCE chain\nAn AI coding agent materially accelerated closed-source exploit research: A Security reports finding and exploiting Zoom\u2019s annotation parser in under 24 hours and fewer than 20 prompts. A crafted 745-byte packet reaches a 128-byte stack buffer via an unchecked count; Zoom corroborates CVE-2026-53413/414/415 and patched affected clients. The RCE demo is macOS and the AI-speed claim is first-party, but shared code makes patching urgent.\n\n\u2728 MazeRunner branches through black-box pentests\nThree LLM agents preserve clues, revise failed actions, and switch attack branches instead of following one linear pentest plan. On ten recent HTB targets under a 20M-token cap, MazeRunner completed 47.7% of subtasks versus 36.2% for PentestGPT-V2, reached user access on six targets, and root on two; both baselines reached user on two and root on none. Human approval is still needed for destructive live actions.\n\n#BlueTeam@bhhub\n\n\u2728 GraphRAG makes threat hunts survive IOC rotation\nKnowledge-graph retrieval pushes AI-generated hunt plans beyond disposable indicators toward behaviors and techniques. In an APT28 case, GraphRAG preserved 100% of detections after every IP, domain, and hash was rotated; naive RAG preserved 29%. Nine reports from four vendors showed the same direction, but prompt wording mattered nearly as much and the corpus is small. Evaluate detection durability, not rule count.", "creation_timestamp": "2026-08-17T08:00:03.938053Z"} https://vulnerability.circl.lu/sighting/dda4a57a-f7d3-4aac-a45b-61be764e2a75/export Mon, 17 Aug 2026 08:00:03 +0000 https://vulnerability.circl.lu/sighting/d163f553-3ec5-4420-b707-b8b7174e8a4f/export {"uuid": "d163f553-3ec5-4420-b707-b8b7174e8a4f", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "published-proof-of-concept", "source": "Telegram/unEmbzqVXcDJ2NUSLzJGWmZlx2cZxv-x1ppCpJ4_UfnK4G4", "content": "", "creation_timestamp": "2026-08-16T00:00:45.811124Z"} {"uuid": "d163f553-3ec5-4420-b707-b8b7174e8a4f", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "published-proof-of-concept", "source": "Telegram/unEmbzqVXcDJ2NUSLzJGWmZlx2cZxv-x1ppCpJ4_UfnK4G4", "content": "", "creation_timestamp": "2026-08-16T00:00:45.811124Z"} https://vulnerability.circl.lu/sighting/d163f553-3ec5-4420-b707-b8b7174e8a4f/export Sun, 16 Aug 2026 00:00:45 +0000 https://vulnerability.circl.lu/sighting/ade08202-3931-4f10-a8f3-a90e95b9b4b8/export {"uuid": "ade08202-3931-4f10-a8f3-a90e95b9b4b8", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "published-proof-of-concept", "source": "Telegram/RUan3U3Ixa6Ril5j2wkYlj3mZcQ0vDhZx_wVxjae8ueeUnE", "content": "", "creation_timestamp": "2026-08-16T00:00:45.455801Z"} {"uuid": "ade08202-3931-4f10-a8f3-a90e95b9b4b8", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "published-proof-of-concept", "source": "Telegram/RUan3U3Ixa6Ril5j2wkYlj3mZcQ0vDhZx_wVxjae8ueeUnE", "content": "", "creation_timestamp": "2026-08-16T00:00:45.455801Z"} https://vulnerability.circl.lu/sighting/ade08202-3931-4f10-a8f3-a90e95b9b4b8/export Sun, 16 Aug 2026 00:00:45 +0000 https://vulnerability.circl.lu/sighting/34a5b208-aa06-4cfa-99ce-26cab2682d9e/export {"uuid": "34a5b208-aa06-4cfa-99ce-26cab2682d9e", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "Telegram/unEmbzqVXcDJ2NUSLzJGWmZlx2cZxv-x1ppCpJ4_UfnK4G4", "content": "", "creation_timestamp": "2026-08-15T08:00:05.881934Z"} {"uuid": "34a5b208-aa06-4cfa-99ce-26cab2682d9e", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "Telegram/unEmbzqVXcDJ2NUSLzJGWmZlx2cZxv-x1ppCpJ4_UfnK4G4", "content": "", "creation_timestamp": "2026-08-15T08:00:05.881934Z"} https://vulnerability.circl.lu/sighting/34a5b208-aa06-4cfa-99ce-26cab2682d9e/export Sat, 15 Aug 2026 08:00:05 +0000 https://vulnerability.circl.lu/sighting/dd697dbf-241a-4f93-b56b-01659700dc85/export {"uuid": "dd697dbf-241a-4f93-b56b-01659700dc85", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "Telegram/RUan3U3Ixa6Ril5j2wkYlj3mZcQ0vDhZx_wVxjae8ueeUnE", "content": "", "creation_timestamp": "2026-08-15T08:00:05.743360Z"} {"uuid": "dd697dbf-241a-4f93-b56b-01659700dc85", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "Telegram/RUan3U3Ixa6Ril5j2wkYlj3mZcQ0vDhZx_wVxjae8ueeUnE", "content": "", "creation_timestamp": "2026-08-15T08:00:05.743360Z"} https://vulnerability.circl.lu/sighting/dd697dbf-241a-4f93-b56b-01659700dc85/export Sat, 15 Aug 2026 08:00:05 +0000 https://vulnerability.circl.lu/sighting/27047ce4-c31d-47a5-81f3-5d963287927f/export {"uuid": "27047ce4-c31d-47a5-81f3-5d963287927f", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://poliverso.org/objects/0477a01e-f489df32-e09466c2552dd19d", "content": "This Week in Security: BugTraq, AI Hacks, and Being Dumb On Planes\nAfter a multi-year hiatus, the venerable BugTraq mailing list is back!\nFor decades, BugTraq was the place where vulnerabilities were disclosed, from the early days when nearly all vendors viewed all security research as a hostile force, through to the modern era of working with vendors to coordinate disclosing bugs. With the rise of bug bounty programs and other social changes, the mailing list slowly died: what started in 1993 ended in 2021 is returning. The new maintainer, Jonathan Brossard, says in his announcement \u201cThe mission is unchanged: full disclosure, researcher-first, no corporate filter.\u201d\nDon\u2019t Be Dumb on Planes\nIn the unlikely event anyone here needs to be told: Don\u2019t do dumb things on planes.\nIt seems that someone coming home from from the DEF CON hacker conference in Las Vegas decided to mess with the in-plane WiFi, and is likely now in the \u201cfind out\u201d phase of doing something dumb. There hasn\u2019t been any public followup beyond the original reports: a passenger on a Delta flight leaving Las Vegas brought up a fake WiFi hotspot named \u201cDelta WiFi FAST\u201d to trick other passengers into connecting, and attempted to disable the in-flight WiFi using a denial of service attack.\nThe recorded messages from the pilots mention that the legitimate WiFi network was \u201cjammed\u201d, though this is unlikely technically correct. While any radio communications can be jammed by transmitting interference, most WiFi networks are susceptible to several basic denial of service attacks due to oversights in the WiFi standards design; by injecting WiFi management packets that tell the clients that a connection to the access point is no longer valid, and the device will attempt to reconnect.\nBy spamming the these disconnect packets continually, the network becomes essentially unusable, but unlike jamming, the interference is caused entirely by legitimate packets causing expected behavior. More recent extensions to the WiFi standards mitigate some of the most trivial attacks by adding additional validation that the packets were actually sent by the access point, but very few networks and clients use the newer options.\nBy rendering the legitimate network useless, the attackers hoped that users would select their fake network instead. None of the reports have included detailed information, but the fake network would likely have prompted the user to enter credentials for another site, such as Google or Facebook. The commercial hotspot model is particularly vulnerable to these sorts of attacks: many hotels, airlines, and businesses run an open, unauthenticated access point, and count on a web page the user must click through to gain access. Users are accustomed to clicking through web pages and providing room or booking numbers, names, or other credentials. Since there is no way to authenticate an open WiFi network, creating a look-alike network can be easy.\nIn lieu of additional official news, Youtuber [ThioJoe] tracked down some information, or the lack thereof, looking for further clarification; it appears that, contrary to some of the more excited online reports, law enforcement was not present when the plane landed, and it\u2019s unclear if there was even an attempt to kick users from the legitimate WiFi network or a sustained \u201cjamming\u201d or denial-of-service. In an official statement from Delta, they say that no in-flight systems were at risk, which is as it should be: customer-facing WiFi is unlikely to be interfaced with airplane flight systems, and even a true jamming attack generating radio interference would be in the WiFi spectrum where radio noise is already expected.\nBut we repeat: an airplane is an exquisitely bad location for illegal shenanigans.\nSteam Hardware Orders Leaked\nThe European handler of Steam hardware orders, CEVA, was breached in July, exposing customer order information such as phone numbers, and email addresses.\nSteam account details were not compromised, but the order information is plenty to arm scammers with enough information to make a convincing phishing attempt. Valve is already explicitly warning impacted customers to expect fake messages pertaining to their orders.\nNo additional information has been available on how the attack compromised CEVA. It appears this only impacts Steam hardware preorders in Europe.\nZoom Exploit\nZoom, the meeting software, has patched three serious vulnerabilities that would allow members of a call to execute arbitrary code on other members of the call.\nCVE-2026-53413 looks to be a typical buffer overflow, where the length of incoming data is not checked properly and overruns the memory space available for it, crashing the Zoom client or potentially allowing code to be executed. CVE-2026-53414 is a closely related buffer size bug, where the buffer allocated is too small for the read operation, crashing the Zoom client. Finally, CVE-2026-53415 is a use-after-free bug, where memory is referenced after it has been released, allowing code execution. The researchers have branded the three vulnerabilities as \u201cZoomsday\u201d.\nThe bad news is that Zoom is a high-profile target for business compromise. All platforms are affected, so if you\u2019re on Zoom, time to update before the next call!\nAI Agent Hacks a Reservation System\nThe Australian Broadcasting Corporation reports on an OpenClaw agent finding flaws in the booking system of a gym when asked to make a reservation. Details are relatively thin, with no complete transcript of the interaction with the OpenClaw agent, but supposedly the agent found that the easiest way to make a booking in the full class was to explore the API of the booking site, discovering that the reservation API was protected, but the cancellation API was not.\nWithin minutes the Claude-backed agent had booked a gym appointment and pushed the user to the top of the list by cancelling other customers. When asked to undo its work, the agent contritely stated that the insertion API was protected and it was impossible, and that it should have performed a test run first before removing users from a live web service.\nWith so few details available it\u2019s difficult to ascertain how accurate the report is. Claude has certainly been caught overstepping bounds recently, and the AI frontier companies seem happy for the press of declaring they accidentally hacked other companies, so it all seems plausible. This does nothing but increase the confusion about legal liability.\nUnder almost any jurisdiction, deliberately hacking the reservation system of a company to jump the line would be considered illegal, but when an agent does it, nobody seems concerned, including the user who launched the request: per the ABC article, \u201cIt\u2019s not the end of the world, so I didn\u2019t beat myself up about it, but it certainly was a warning signal to use it responsibly.\u201d\nUnited States to Authorize Private Companies for Cyber Operations\nAfter brain-draining existing government cyber agencies like CISA, this week the US Government issued a memo authorizing the National Coordination Center, part of the Department of Homeland Security to use private companies to perform cyber security responses on behalf of the United States.\nThe memo requires companies to obey the US Constitution and federal law, and requires a one million dollar bond for each company, but still represents a massive change in the posture of the US government with regards to cyber security. Previously, cyber incidents and responses would often by handled by CISA, the US Army Cyber Command, or the NSA. Allowing private companies to respond on behalf of the country feels almost like a return to letters of marque authorizing piracy and freebooting in the 1800s.\nLiteLLM Supply Chain\nSome weeks ago we mentioned the compromise of the open source supply chain vulnerability scanner, Trivy. Infected with a supply chain worm which steals credentials and infects every package and build it can now access, the worm has continued to spread, now compromising LiteLLM which is an AI proxy and gateway package used by thousands of enterprises.\nLiteLLM was infected because they utilized Trivy as part of the automated build workflow, and the compromise was not detected. Once the worm had access to the LiteLLM publishing credentials, it pushed new versions of the LiteLLM packages, of course infected with the same credential, authentication token, and cryptocurrency stealing code. Like the NPM node.js based worms, this code triggered using the Python startup hooks, causing it to execute as soon as Python indexed available libraries, even if the infected libraries themselves were never used.\nThe firm Hudson Rock was able to obtain the archive of dumped credentials and data, saying it was 153 GB of compressed content. Trivy, in turn, was compromised in March of 2026 after a misconfigured GitHub workflow allowed a pull request to extract authentication keys. The keys were not fully disabled, and the attacker returned weeks later to infect over 50 Trivy packages and workflows.\nAnalyzing the contents of the stolen data, the list of credentials stolen is staggering, including GitHub and GitLab credentials for Boeing, Orange Telecom, Roku, and multiple government agencies and labs, as well as Slack credentials, SSH keys, and cloud computing access. Hudson Rock has created a site to look up domain to see if your company has been impacted.\nSupply chain compromises continue to spread and impact thousands of packages, but the final goal clearly isn\u2019t to simply spread between packages. When high profile heavily used packages are compromised, the stolen tokens will be used to breach the affected companies sooner or later. The various package ecosystems are still struggling to find solutions to poisoned packages that don\u2019t break existing automation processes, and until those are solved, we can expect continual hacks like these to succeed.\nHacking Solar Inverters, Part 2\nRecently, unauthenticated flaws over radio were discovered in solar inverter systems popular as \u201cpatio solar\u201d installs in Europe. this week, SaiFlow details vulnerabilities in the REST API of FIMER inverters.\nInverters convert between DC and AC current, and optionally between DC and DC for hybrid systems. They are a core part of energy systems that interface solar, battery, and power grids. SaiFlow targeted FIMER inverters because, in their own words, \u201cbatteries catch fire\u201d if the system goes wrong badly enough. They found that the inverters are made of several interconnected components, but at the heart is an embedded Linux system built with buildroot, a system framework similar to traditional distributions or a framework like OpenWRT. Due to a misconfiguration in the nginx webserver, API requests can be sent with no authentication at all, and one of the endpoints allows direct injection of commands over a proprietary protocol.\nThe proprietary protocol, Aurora, predates internet connectivity on the devices, and lacks any modern protection or authentication. SaiFlow discovered that Aurora uses a six-digit PIN, which would be simple to brute force if the function to read the PIN from flash weren\u2019t also available, with no authentication. And the PIN is disabled by default.\nSince Aurora was designed as an internal low-level protocol for controlling the hardware, it has access to critical safety features and the ability to override them. The most significant finding was the ability to enable feeding power to the grid, even if the grid is detected as being disabled. This could damage the inverter itself as it tries to power the whole neighborhood, but worse, like a generator plugged in incorrectly, it could electrocute repair workers handling the power lines. SaiFlow points out this could be enabled fully remotely through the unauthenticated web interface. Other exposed commands allow writing arbitrary data to flash, corrupting the unit, changing the country standards, generating incorrect power exported to the grid, and control over charge and discharge rates and battery charge control.\nVulnerabilities in infrastructure components can be difficult to fix, and SaiFlow reports they have received no meaningful response from FIMER months after reporting them. Hopefully effective patches can be developed soon. \nhackaday.com/2026/08/14/this-w\u2026", "creation_timestamp": "2026-08-14T14:24:46.314813Z"} {"uuid": "27047ce4-c31d-47a5-81f3-5d963287927f", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://poliverso.org/objects/0477a01e-f489df32-e09466c2552dd19d", "content": "This Week in Security: BugTraq, AI Hacks, and Being Dumb On Planes\nAfter a multi-year hiatus, the venerable BugTraq mailing list is back!\nFor decades, BugTraq was the place where vulnerabilities were disclosed, from the early days when nearly all vendors viewed all security research as a hostile force, through to the modern era of working with vendors to coordinate disclosing bugs. With the rise of bug bounty programs and other social changes, the mailing list slowly died: what started in 1993 ended in 2021 is returning. The new maintainer, Jonathan Brossard, says in his announcement \u201cThe mission is unchanged: full disclosure, researcher-first, no corporate filter.\u201d\nDon\u2019t Be Dumb on Planes\nIn the unlikely event anyone here needs to be told: Don\u2019t do dumb things on planes.\nIt seems that someone coming home from from the DEF CON hacker conference in Las Vegas decided to mess with the in-plane WiFi, and is likely now in the \u201cfind out\u201d phase of doing something dumb. There hasn\u2019t been any public followup beyond the original reports: a passenger on a Delta flight leaving Las Vegas brought up a fake WiFi hotspot named \u201cDelta WiFi FAST\u201d to trick other passengers into connecting, and attempted to disable the in-flight WiFi using a denial of service attack.\nThe recorded messages from the pilots mention that the legitimate WiFi network was \u201cjammed\u201d, though this is unlikely technically correct. While any radio communications can be jammed by transmitting interference, most WiFi networks are susceptible to several basic denial of service attacks due to oversights in the WiFi standards design; by injecting WiFi management packets that tell the clients that a connection to the access point is no longer valid, and the device will attempt to reconnect.\nBy spamming the these disconnect packets continually, the network becomes essentially unusable, but unlike jamming, the interference is caused entirely by legitimate packets causing expected behavior. More recent extensions to the WiFi standards mitigate some of the most trivial attacks by adding additional validation that the packets were actually sent by the access point, but very few networks and clients use the newer options.\nBy rendering the legitimate network useless, the attackers hoped that users would select their fake network instead. None of the reports have included detailed information, but the fake network would likely have prompted the user to enter credentials for another site, such as Google or Facebook. The commercial hotspot model is particularly vulnerable to these sorts of attacks: many hotels, airlines, and businesses run an open, unauthenticated access point, and count on a web page the user must click through to gain access. Users are accustomed to clicking through web pages and providing room or booking numbers, names, or other credentials. Since there is no way to authenticate an open WiFi network, creating a look-alike network can be easy.\nIn lieu of additional official news, Youtuber [ThioJoe] tracked down some information, or the lack thereof, looking for further clarification; it appears that, contrary to some of the more excited online reports, law enforcement was not present when the plane landed, and it\u2019s unclear if there was even an attempt to kick users from the legitimate WiFi network or a sustained \u201cjamming\u201d or denial-of-service. In an official statement from Delta, they say that no in-flight systems were at risk, which is as it should be: customer-facing WiFi is unlikely to be interfaced with airplane flight systems, and even a true jamming attack generating radio interference would be in the WiFi spectrum where radio noise is already expected.\nBut we repeat: an airplane is an exquisitely bad location for illegal shenanigans.\nSteam Hardware Orders Leaked\nThe European handler of Steam hardware orders, CEVA, was breached in July, exposing customer order information such as phone numbers, and email addresses.\nSteam account details were not compromised, but the order information is plenty to arm scammers with enough information to make a convincing phishing attempt. Valve is already explicitly warning impacted customers to expect fake messages pertaining to their orders.\nNo additional information has been available on how the attack compromised CEVA. It appears this only impacts Steam hardware preorders in Europe.\nZoom Exploit\nZoom, the meeting software, has patched three serious vulnerabilities that would allow members of a call to execute arbitrary code on other members of the call.\nCVE-2026-53413 looks to be a typical buffer overflow, where the length of incoming data is not checked properly and overruns the memory space available for it, crashing the Zoom client or potentially allowing code to be executed. CVE-2026-53414 is a closely related buffer size bug, where the buffer allocated is too small for the read operation, crashing the Zoom client. Finally, CVE-2026-53415 is a use-after-free bug, where memory is referenced after it has been released, allowing code execution. The researchers have branded the three vulnerabilities as \u201cZoomsday\u201d.\nThe bad news is that Zoom is a high-profile target for business compromise. All platforms are affected, so if you\u2019re on Zoom, time to update before the next call!\nAI Agent Hacks a Reservation System\nThe Australian Broadcasting Corporation reports on an OpenClaw agent finding flaws in the booking system of a gym when asked to make a reservation. Details are relatively thin, with no complete transcript of the interaction with the OpenClaw agent, but supposedly the agent found that the easiest way to make a booking in the full class was to explore the API of the booking site, discovering that the reservation API was protected, but the cancellation API was not.\nWithin minutes the Claude-backed agent had booked a gym appointment and pushed the user to the top of the list by cancelling other customers. When asked to undo its work, the agent contritely stated that the insertion API was protected and it was impossible, and that it should have performed a test run first before removing users from a live web service.\nWith so few details available it\u2019s difficult to ascertain how accurate the report is. Claude has certainly been caught overstepping bounds recently, and the AI frontier companies seem happy for the press of declaring they accidentally hacked other companies, so it all seems plausible. This does nothing but increase the confusion about legal liability.\nUnder almost any jurisdiction, deliberately hacking the reservation system of a company to jump the line would be considered illegal, but when an agent does it, nobody seems concerned, including the user who launched the request: per the ABC article, \u201cIt\u2019s not the end of the world, so I didn\u2019t beat myself up about it, but it certainly was a warning signal to use it responsibly.\u201d\nUnited States to Authorize Private Companies for Cyber Operations\nAfter brain-draining existing government cyber agencies like CISA, this week the US Government issued a memo authorizing the National Coordination Center, part of the Department of Homeland Security to use private companies to perform cyber security responses on behalf of the United States.\nThe memo requires companies to obey the US Constitution and federal law, and requires a one million dollar bond for each company, but still represents a massive change in the posture of the US government with regards to cyber security. Previously, cyber incidents and responses would often by handled by CISA, the US Army Cyber Command, or the NSA. Allowing private companies to respond on behalf of the country feels almost like a return to letters of marque authorizing piracy and freebooting in the 1800s.\nLiteLLM Supply Chain\nSome weeks ago we mentioned the compromise of the open source supply chain vulnerability scanner, Trivy. Infected with a supply chain worm which steals credentials and infects every package and build it can now access, the worm has continued to spread, now compromising LiteLLM which is an AI proxy and gateway package used by thousands of enterprises.\nLiteLLM was infected because they utilized Trivy as part of the automated build workflow, and the compromise was not detected. Once the worm had access to the LiteLLM publishing credentials, it pushed new versions of the LiteLLM packages, of course infected with the same credential, authentication token, and cryptocurrency stealing code. Like the NPM node.js based worms, this code triggered using the Python startup hooks, causing it to execute as soon as Python indexed available libraries, even if the infected libraries themselves were never used.\nThe firm Hudson Rock was able to obtain the archive of dumped credentials and data, saying it was 153 GB of compressed content. Trivy, in turn, was compromised in March of 2026 after a misconfigured GitHub workflow allowed a pull request to extract authentication keys. The keys were not fully disabled, and the attacker returned weeks later to infect over 50 Trivy packages and workflows.\nAnalyzing the contents of the stolen data, the list of credentials stolen is staggering, including GitHub and GitLab credentials for Boeing, Orange Telecom, Roku, and multiple government agencies and labs, as well as Slack credentials, SSH keys, and cloud computing access. Hudson Rock has created a site to look up domain to see if your company has been impacted.\nSupply chain compromises continue to spread and impact thousands of packages, but the final goal clearly isn\u2019t to simply spread between packages. When high profile heavily used packages are compromised, the stolen tokens will be used to breach the affected companies sooner or later. The various package ecosystems are still struggling to find solutions to poisoned packages that don\u2019t break existing automation processes, and until those are solved, we can expect continual hacks like these to succeed.\nHacking Solar Inverters, Part 2\nRecently, unauthenticated flaws over radio were discovered in solar inverter systems popular as \u201cpatio solar\u201d installs in Europe. this week, SaiFlow details vulnerabilities in the REST API of FIMER inverters.\nInverters convert between DC and AC current, and optionally between DC and DC for hybrid systems. They are a core part of energy systems that interface solar, battery, and power grids. SaiFlow targeted FIMER inverters because, in their own words, \u201cbatteries catch fire\u201d if the system goes wrong badly enough. They found that the inverters are made of several interconnected components, but at the heart is an embedded Linux system built with buildroot, a system framework similar to traditional distributions or a framework like OpenWRT. Due to a misconfiguration in the nginx webserver, API requests can be sent with no authentication at all, and one of the endpoints allows direct injection of commands over a proprietary protocol.\nThe proprietary protocol, Aurora, predates internet connectivity on the devices, and lacks any modern protection or authentication. SaiFlow discovered that Aurora uses a six-digit PIN, which would be simple to brute force if the function to read the PIN from flash weren\u2019t also available, with no authentication. And the PIN is disabled by default.\nSince Aurora was designed as an internal low-level protocol for controlling the hardware, it has access to critical safety features and the ability to override them. The most significant finding was the ability to enable feeding power to the grid, even if the grid is detected as being disabled. This could damage the inverter itself as it tries to power the whole neighborhood, but worse, like a generator plugged in incorrectly, it could electrocute repair workers handling the power lines. SaiFlow points out this could be enabled fully remotely through the unauthenticated web interface. Other exposed commands allow writing arbitrary data to flash, corrupting the unit, changing the country standards, generating incorrect power exported to the grid, and control over charge and discharge rates and battery charge control.\nVulnerabilities in infrastructure components can be difficult to fix, and SaiFlow reports they have received no meaningful response from FIMER months after reporting them. Hopefully effective patches can be developed soon. \nhackaday.com/2026/08/14/this-w\u2026", "creation_timestamp": "2026-08-14T14:24:46.314813Z"} https://vulnerability.circl.lu/sighting/27047ce4-c31d-47a5-81f3-5d963287927f/export Fri, 14 Aug 2026 14:24:46 +0000 https://vulnerability.circl.lu/sighting/ca8f6c92-bcf5-471b-bf03-39430297846d/export {"uuid": "ca8f6c92-bcf5-471b-bf03-39430297846d", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://bsky.app/profile/whisprnews.bsky.social/post/3msy3k3e5te2l", "content": "\ud83d\udea8 'Zoomsday': A Security destapa dos fallos zero-click en Zoom (CVE-2026-53413 y CVE-2026-53415, severidad 8.3) que dan control del dispositivo sin un solo click. En riesgo: sesiones de intercambio y software de billetera. Parchea ya: 7.1.5 / 7.0.6.", "creation_timestamp": "2026-08-13T16:41:29.355438Z"} {"uuid": "ca8f6c92-bcf5-471b-bf03-39430297846d", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://bsky.app/profile/whisprnews.bsky.social/post/3msy3k3e5te2l", "content": "\ud83d\udea8 'Zoomsday': A Security destapa dos fallos zero-click en Zoom (CVE-2026-53413 y CVE-2026-53415, severidad 8.3) que dan control del dispositivo sin un solo click. En riesgo: sesiones de intercambio y software de billetera. Parchea ya: 7.1.5 / 7.0.6.", "creation_timestamp": "2026-08-13T16:41:29.355438Z"} https://vulnerability.circl.lu/sighting/ca8f6c92-bcf5-471b-bf03-39430297846d/export Thu, 13 Aug 2026 16:41:29 +0000 https://vulnerability.circl.lu/sighting/177832e1-de9b-4a0c-9344-e88dd03fa466/export {"uuid": "177832e1-de9b-4a0c-9344-e88dd03fa466", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "86ecb4e1-bb32-44d5-9f39-8a4673af8385", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://www.ncsc.nl/alerts/kwetsbaarheden-in-zoom-clients-en-zoom-vdi-client-software", "content": "", "creation_timestamp": "2026-08-13T12:15:03.833258Z"} {"uuid": "177832e1-de9b-4a0c-9344-e88dd03fa466", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "86ecb4e1-bb32-44d5-9f39-8a4673af8385", "vulnerability": "CVE-2026-53413", "type": "seen", "source": "https://www.ncsc.nl/alerts/kwetsbaarheden-in-zoom-clients-en-zoom-vdi-client-software", "content": "", "creation_timestamp": "2026-08-13T12:15:03.833258Z"} https://vulnerability.circl.lu/sighting/177832e1-de9b-4a0c-9344-e88dd03fa466/export Thu, 13 Aug 2026 12:15:03 +0000