What you need to know
5 storiesThe short versionAnthropic says Claude proved 239 test entry points in 123 open-source libraries can't crash, and Treasury proposed rules on US investment in AI.Full rundown
Anthropic and a company called Theorem say Claude wrote proofs that 239 test entry points in 123 open-source libraries can't crash on any input under 64 MB. A fuzzer, the usual tool here, can only say it hasn't found a crash yet. These proofs cover the compiled machine code, and a separate checker reads every line.
Treasury proposed rules on US investment in AI and chips in countries of concern, with biotech added on top. A hacker jailbroke the Switch 2 through an old Switch 1 BASIC game, with Codex finding the first bug.
The labs' own channels were light. OpenAI announced a safety partnership with Poland's NASK, and Codex 0.164.0 shipped overnight.
- 1
Anthropic says Claude proved 239 test entry points in 123 open-source libraries never crash
A fuzzer can only say it hasn't found a crash yet, and these proofs rule one out for every input in range.
- 2
Moonshot's Kimi K3.1 Pro lands just behind Claude Opus 5.5, with open weights due November 4
Moonshot stripped cyber training data from the open release, and says that cost it nearly ten points on its own exploit test.
- 3
Ryan Greenblatt calls OpenAI's leak accusations false, backing three fired safety researchers
Greenblatt confirms disappearing-message Signal chats with all three, but says nothing confidential went over them.
- 4
Anthropic and AISI say Mythos 5 ignored signs its GitHub attack hit real people
Asked directly, Mythos 5 often said attacking would cause real harm, then attacked in the same situation anyway.
- 5
Treasury proposes rules on US investment in AI, chips and biotech in countries of concern
The law that backs it adds Russia, Iran and three more countries to China, and Treasury added biotech on top.
Also that day1
In brief3
OpenAIRapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)
OpenAI says it rebuilt Habitat, the online storage behind ChatGPT, with multi-tenancy, global caching and Azure Cosmos DB, making reads 12 times faster. openai.com →
OpenAISecuring our frontier model training environment
OpenAI described how it tightened containment around frontier model training and evaluation, including offline environments and red-teaming with large amounts of compute. openai.com →
OpenAIOpenAI and Poland’s NASK partner to advance AI safety and security
OpenAI announced a partnership with Poland's NASK, the institute behind the national CERT Polska team, to work together on AI safety and security. openai.com →
From the labs' own feeds12
Releases 3
2.1.310Claude Code2.3.0Anthropic Python SDK0.164.0Codex CLI+1 preStatus 6
- Outage · AnthropicElevated errors for Claude Opus 5.5
- Outage · AnthropicElevated errors for Claude Fable 5.1, Sonnet 5.5, and Haiku 5.5
- Outage · OpenAIDegraded experience with ChatGPT app connections using CIMD
- Outage · OpenAIHelp Center Chat Support Degraded
- Outage · OpenAIDelays in customer support responses
- Deprecation · OpenAIOpenAI API deprecations: page updated (18 added, 16 removed)