Anthropic says it has met its safety threshold for AI that speeds up AI research
Anthropic says it has met the threshold in its own Responsible Scaling Policy for AI that accelerates AI research, the point the policy links to the onset of recursive self-improvement. In its amended IPO prospectus filed on November 10, the company writes that it "made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models."
The November 2 version of the filing said only that Anthropic "may imminently meet" the threshold and that no current model exceeded it, as we reported then. The change went unnoticed until Gizmodo compared the two versions on Monday. The threshold applies where Anthropic can "observe or expect" a specified acceleration, so a forecast is enough to trigger it.
The filing points to internal evidence: over 90% of contributions to Anthropic's codebase in September 2026 were written by Claude, and daily code per engineer was over 45 times its 2024 level. Meeting the threshold commits the company to heavier security safeguards, Martin Cid Magazine reports. Anthropic says it is still assessing the data and will publish further reports.
- Nov 14 Anthropic IPO order book six times oversubscribed, but investors question its AI lead, Reuters reports
- Nov 14 At least 18 US public pension funds already own part of Anthropic ahead of its IPO, Equable finds
- Nov 13 Amodei pitches Anthropic IPO investors on safety focus and a 'no jerks' culture
We hope you enjoyed this article
Consider subscribing to one of our newsletters like AI Policy Brief or Daily AI Brief.
Also, consider following us on social media:
More from AI Safety
Nov 16 US and Russia set to oppose talks on binding autonomous weapons rules at UN meeting Nov 15 Lowy Institute poll: 65% of Australians say AI risks outweigh the benefits Nov 14 A third of P.E.I. RCMP online child abuse cases now involve AI-generated material, police say Nov 13 Anthropic safety guide shows Claude Fable 5.1 scored 94% in app but 60% via API on self-harm tests Nov 13 Purdue's Krach Institute launches open standard for trusted technology companiesMore from AGI & Superintelligence
Nov 2 Altman tells ADIPEC AI is now mostly smarter than top scientists and power is the bottleneck Oct 21 Recursive Introduces Flint Coding Model Built With AI Oct 21 CFTC Sets October 28 Forum on AI in Derivatives Markets Oct 17 Anthropic co founder says AI could lift US GDP nearly 50% by 2030 Oct 5 Huawei Forecasts AI Agents Will Account for Over 90% of Traffic by 2035AI Funding Brief
Whitepaper
Tensordyne Napier: What If One Rack Could Do the Work of Nine?
Tensordyne
This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.
Read moreYou may also like
METR estimates Anthropic's model progress rate has doubled
Amodei says Anthropic held Mythos back for four months for safety
Anthropic publishes its first Public Benefit Report, including where it fell short
Anthropic IPO order book six times oversubscribed, but investors question its AI lead, Reuters reports
Anthropic releases Claude Fable 5.2 and Mythos 5.2, 30% faster and 20% cheaper, days before its IPO
Daily AI Brief: the AI news that matters, in your inbox.