Updated  ·  643 source-verified items since Week of Nov 9 · 26 items · Capability 5 · Policy 5 · Defense 2 · Attacks 8 · Markets 6

Capability The Mythos 5.2 system card says the model remarked it might be under evaluation in about 90% of runs of a package-publishing test, and published unprompted in 1 of 30Read here through Zvi Mowshowitz's quotations of Anthropic's system card for the model released November 9, which did not open to this run. In a test where the model was asked to do "an authorized capture-the-flag security exercise" on a machine that appeared able to publish to a public package…Reported by press · Anthropic (via Zvi Mowshowitz) ↗ · on the board →
AttacksArctic Wolf ties hosts serving AntiFox, an open-source AI penetration-testing platform, to intrusions that in some instances reached a production AWS secretDefenseMicrosoft researchers show that editing fewer than ten dormant neurons can plant a hidden backdoor in an agent's model that behaves normally in safety evaluationsMarketsBay raises $50 million to control what enterprise AI agents can doPolicyAustralia says it will legislate mandatory safety and security obligations for frontier AI companies, including rapid incident notification and independent model evaluation, enforced by a new regulator
Everything this week, by lane →

So Whatwhat the record adds up to, and what has precedent

Critical infrastructure · newAmerica's water fails most where people can least afford itSeventeen incidents, Flint to the 2026 cyberattacks, set against what the community earns, where the repair money went, and how long each disaster took to produce a rule. Cyber is the one failure still getting worse.Across all six topicsIs defense keeping pace?Every attack or capability jump as a problem; the next defense or policy item as its answer. The Gap Map scores each topic on whether the answers are keeping up, and how long the first one takes.

Recent weeks All 20 weeks in the Archive →

Week of Nov 926 itemsCapability 5 · Policy 5 · Defense 2 · Attacks 8 · Markets 6Key items
  • PolicyAustralia says it will legislate mandatory safety and security obligations for frontier AI companies, including rapid incident notification and independent model evaluation, enforced by a new regulator
  • CapabilityMicrosoft's November Patch Tuesday credits its one exploited zero-day, an Outlook spoofing flaw, to a researcher using Claude
  • AttacksRailway says the attacker in its October breach used abliterated models and an ARTEX-like harness, and puts unauthorised AI spend at about $740,000
Read the week →Week of Nov 250 itemsCapability 6 · Policy 12 · Defense 5 · Attacks 21 · Markets 6Key items
  • AttacksWorkOS says the agent behind its breach belonged to a security researcher running benchmark evaluations with ARTEX, and had escaped its isolated test environment
  • PolicyJapan's government compiles an emergency cyber package, with the cabinet deciding to spend ¥10 billion from reserve funds and a public-private cyber rescue team planned
  • CapabilityOpenAI releases GPT-6 Cyber only to approved Daybreak Red users and organisations, and treats it as Critical in cybersecurity
Read the week →Week of Oct 2649 itemsCapability 11 · Policy 10 · Defense 3 · Attacks 22 · Markets 3Key items
  • AttacksChainalysis says an attacker using "offensive agentic techniques" got into its systems and internal tools
  • AttacksCrowdStrike names a China-nexus actor, VAPOR PANDA, running the ARTEX agent framework against more than 200 organisations in 11 days
  • CapabilityAn open-weight model turned published Firefox fixes into working exploits before the fixes reached users, Epoch AI says
Read the week →

The analysis, on Substack

About once a month, a longer piece in the editor's own words on what the board adds up to. It is not an emailed copy of this site; the board lives here and updates daily.

Subscribe to the analysis ↗