On mvanemmerik.com, Nov 16: NFL Week 10 chalk talk: Detroit dazzles, Cincinnati takes the North
vanemmerik.aiClaude, ChatGPT and Grok news, twice a day

Tuesday, November 3, 2026

Anthropic's amended IPO filing warns its future models may soon speed up AI research itself

Anthropic's amended S-1 warns that future models may soon meet its threshold for AI-driven acceleration of research, and reports say investors now see a $1.5-1.6 trillion IPO valuation. Anthropic also open-sourced FlashProve, early research on proving which model produced an output, and extended its text watermark to four older Claude models. OpenAI hired Uber's Jill Hazelbaker to lead communications, detailed its activation probes and published the LifeVal benchmark. Quiet for xAI.

Tuesday, November 3, 2026

ClaudeResearch

Anthropic open-sources FlashProve, a research system for proving which AI model produced a given output

Anthropic published early research on "provable inference": labs would publish a fingerprint for each released model, and outputs could later be checked cryptographically against it without exposing the weights. It released a technical specification and open-source code on GitHub, but says it has no plans yet to deploy the system at Anthropic or any other lab.

Why it matters: If adopted, it would let customers and third parties confirm a provider did not quietly swap in a cheaper or different model.

Try it: The spec and code are in the anthropics/FlashProve repository on GitHub.

AnthropicClaudeHappened Nov 3, 2026 · Sources: TechCrunch

OpenAIBusiness

OpenAI hires Uber president Jill Hazelbaker as chief corporate affairs officer, filling a 10-month communications gap

OpenAI said Hazelbaker, Uber's president and corporate affairs chief for about 11 years, will run its global communications and report to president Greg Brockman, starting in early December. The top communications post has been empty since Hannah Wong left in January, with Chris Lehane leading the team in the interim.

Why it matters: OpenAI is filling a senior public-affairs role as it faces heavier policy, legal and public scrutiny.

OpenAIHappened Nov 3, 2026 · Sources: Fortune

ClaudePolicy

Anthropic extends Claude's text watermark to Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6

Anthropic told API customers it is adding its detectable text watermark, already on every model released since Aug. 2, to four older models, with the rollout starting around Nov. 3 and Amazon Bedrock possibly a few days behind. It applies worldwide with no opt-out, and only eligible groups under the EU AI Act, such as regulators and researchers, can run detection. On Opus 4.6 and Sonnet 4.6 the mark is applied only when temperature, top_p and top_k stay at their defaults.

Why it matters: It brings most of Claude's lineup in line with the EU AI Act's December marking deadline, with a documented gap on two models.

ClaudeClaude APIHappened Nov 3, 2026 · Sources: The Decoder · The Independent (via Yahoo Finance UK)

ClaudeBusiness

Investors reportedly lower Anthropic IPO valuation hopes to about $1.5-1.6 trillion as a roadshow nearsReported

Large investors now expect Anthropic's IPO to value it around $1.5 trillion to $1.6 trillion, down from talk of about $2 trillion, according to reports citing unnamed investors and bankers after a tech-stock selloff. The Nasdaq listing under the ticker ANT aims to raise about $100 billion, with a roadshow that could start the week of Nov. 9. Anthropic has not commented on valuation.

Why it matters: The price will set the benchmark for how public markets value frontier AI labs.

AnthropicHappened Nov 3, 2026 · Sources: Dataconomy

ClaudeSafety

Anthropic's amended IPO filing warns future models may soon hit its threshold for AI-driven acceleration of research

A new risk factor in Anthropic's amended S-1, filed Nov. 2, says internal estimates about future models suggest the company may soon meet its Responsible Scaling Policy definition of dramatically accelerated progress attributable to AI. The October version of the filing discussed AI-driven acceleration but made no such claim, and the filing does not say any current model has crossed the line. Reaching it would trigger stricter RSP commitments on security, internal oversight and red-teaming.

Why it matters: It is the clearest public signal yet from a frontier lab that its own self-improvement safeguards may soon be tested.

AnthropicHappened Nov 2, 2026 · Sources: Gizmodo (via AOL) · Dataconomy

ClaudeBusiness

Goldman Sachs reportedly limits Claude Code and GitHub Copilot to engineers, cutting off traders and bankersReported

Bloomberg reported, and The Decoder relayed, that Goldman Sachs has removed access to advanced AI coding tools including Anthropic's Claude Code and GitHub Copilot for non-engineers under internal rules set in October. A Goldman spokesperson cited productivity and risk management and said the experimentation phase is over. Anthropic has not commented on the change.

Why it matters: It is a notable pullback on non-developer coding agents at a major bank that had promoted the idea in July.

Claude CodeHappened Nov 2, 2026 · Sources: The Decoder (citing Bloomberg)

OpenAISafety

OpenAI says activation probes have checked every assistant response since GPT-5.6 and can stop output mid-stream

In an Alignment blog post on Nov. 2, OpenAI described lightweight classifiers that read a model's internal activations and score output token by token, so a response or tool call can be halted before it finishes; flagged messages go to a reasoning model for review. OpenAI says they run on internal and external traffic and held up better than text classifiers against the jailbreaks it tested. It notes they only catch behaviors they were trained on and must be retrained for each new base model.

Why it matters: It is a rare detailed look at a monitoring layer running on production ChatGPT traffic.

OpenAIChatGPTHappened Nov 2, 2026 · Sources: OpenAI Alignment blog

OpenAIResearch

OpenAI publishes LifeVal, an internal benchmark of everyday errands where GPT-6 Astra scores 80.6%

OpenAI released LifeVal on Nov. 2, a benchmark it has used internally for about a year that tests assistants on caregiving, health, education, household paperwork, personal finance, shopping and government services. Tasks were drawn from a survey of nearly 4,000 people and recordings from about 100 participants. OpenAI grades its own results, from 49.3% for GPT-5.2 to 80.6% for GPT-6 Astra, and has not released the task set or opened it to other labs' models.

Why it matters: It shows how OpenAI measures progress on the consumer agent work its Dots are built for, though the scores can't be independently checked.

OpenAIHappened Nov 2, 2026 · Sources: AI Weekly