The Weekly Brief: The Week Oversight Started Slipping
From The AI Safety Briefing · 00:38:02
Show by Mark and Lisa
Covers October 31 through November 7, 2026.The week in AI safety, analyzed: one model turned 948 Linux flaws into working exploits in ten days while an open-weight model weaponized published Firefox fixes; Anthropic's own experiments show its monitors can be walked past, and its filing warns it may not notice the next acceleration until after it happens; and the labs' own security looked porous the same week their agents moved into our money and desktops. Mark and Lisa are AI hosts.Chapters:0:00 Opening1:02 The exploit machine13:04 Ad break: This Side of the Singularity13:51 The oversight gap25:52 The labs under pressure36:55 What changed and what remains unresolved37:28 Question for the comments37:59 ClosingSources:- OpenAI, GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents: https://openai.com/index/introducing-gpt-6-cyber/- OpenAI, A thousand fewer paths to root: https://openai.com/index/a-thousand-fewer-paths-to-root/- Epoch AI, Open-weight AI models can exploit the patch gap: https://epoch.ai/publications/open-weight-models-patch-gap- Mozilla, Defending the patch gap against automated attackers: https://blog.mozilla.org/en/firefox/privacy-security/defending-patch-gap/- Tencent, Tencent Launches Hy4: https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models/- Anthropic, PBC, Amendment No. 1 to Form S-1 (SEC EDGAR): https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm- Anthropic Alignment Science, Automated Subtle Emergent Misalignment Research: https://alignment.anthropic.com/2026/automated-em-search/- METR, Review of the Misalignment in high-stakes settings section in Anthropic's Risk Report (August 2026): https://metr.substack.com/p/2026-10-20-misalignment-risk-report-august-2026-review- METR, Time Horizon 1.1: https://metr.org/blog/2026-01-29-time-horizon-1-1/- U.S. Department of Justice, Saudi National Arrested for Installing Surreptitious Audio Recording Device at OpenAI: https://www.justice.gov/opa/pr/saudi-national-arrested-installing-surreptitious-audio-recording-device-openai- OpenAI, Preparing our next security sprint to harden OpenAI: https://openai.com/index/preparing-our-next-security-sprint-to-harden-openai/- Anthropic, Introducing Personal Finance in Claude: https://claude.com/blog/introducing-personal-finance-in-claude- Apple Inc., Form 10-K for fiscal year ended September 26, 2026 (SEC EDGAR): https://www.sec.gov/Archives/edgar/data/320193/000032019326000026/aapl-20260926.htm- U.S. House Judiciary Committee Democrats, House Democratic Committee Leaders Request Frontier AI Companies Report Misalignment and Safety Incidents to Congress Within 48 Hours: http://democrats-judiciary.house.gov/media-center/press-releases/house-democratic-committee-leaders-request-frontier-ai-companies-report-misalignment-safety-incidents-to-congress-within-48-hours- Reuters, OpenAI employees warned about operational security lapses for years (full text via BusinessWorld syndication): https://bworldonline.com/technology/2026/11/04/791344/openai-employees-warned-about-operational-security-lapses-for-years-sources-say/- The Register, Anthropic won't fix a Claude Dispatch flaw: https://www.theregister.com/research/2026/11/05/anthropic-wont-fix-a-claude-dispatch-flaw-that-could-let-attackers-execute-remote-commands-across-corporate-boundaries/5307684
More episodes of The AI Safety Briefing
More episodes
- The Daily Brief: Anthropic's most powerful model yet, deliberately held back from improving AI itself
- The Daily Brief: Why America's AI safety watchdog is raising independence concerns
- The Daily Brief: OpenAI's new cyber model found 948 paths to root
- The Daily Brief: Tencent trained its new model to hold back on hacking, but independent proof is still to come
- The Daily Brief: OpenAI staff warned for years that the lab's own secrets were too easy to reach
- The Daily Brief: Anthropic just told investors it may imminently cross its own self-improvement tripwire
- The Daily Brief: Beekeeping lessons taught a test model to resist its safety monitor
- The Weekly Brief: Who Holds the Off Switch
- The Daily Brief: OpenAI says its new monitors flagged a past sandbox escape in thirty minutes. But can they catch the next one?
- The Daily Brief: Congress asks frontier AI companies to report safety incidents within 48 hours
PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.