Anthropic Commits 5% of Its Shares to a Public Benefit Fund
Anthropic announced the Anthropic Public Benefit Fund in a Sunday blog post. It pledges 5% of the company’s shares with immediate effect, subject to stockholder approval, and adds 2% at the board’s discretion for every 50% rise in company value.
The fund can back nonprofits and direct distributions, including work on AI labor-market disruption. It cannot be used for lobbying or political donations. Al Jazeera notes Anthropic is preparing a possible record IPO before the end of 2026, with analysts forecasting a valuation above $2 trillion.
Claude Designs Protein Binders for Scorpion Neurotoxins
Anthropic says an internal Claude research model, working mostly autonomously inside Claude Science over 64 hours, generated 318,484 protein designs aimed at five scorpion neurotoxins. A set of 211 was tested by Adaptyv Bio and Twist Bioscience, and 17 met Anthropic’s binding criteria. The strongest Cn2 binders hit the single-digit nanomolar range.
Anthropic calls it a research demonstration, not an antivenom. Neutralization, safety and efficacy are unproven, and testing used recombinant toxins. All 211 designs and the binding data are being released for other researchers.
Claude Code 2.1.304 and 2.1.305 Land
Version 2.1.304 extends guided auto-mode setup (/auto-mode-setup) to Bedrock, Vertex and Foundry, adds a /claude-api managed-agents-secure check against Anthropic’s security guide, and gives Code Review Quick, Standard and Deep levels.
Version 2.1.305 adds sandbox.allowNestedUserNamespaces for Linux sandboxed commands, with a warning that it lets those commands get around the sandbox’s file protections. It also fixes retry and tool-search bugs and adds an unattended-retry usage-limit setting.
Inference Hooks Leave Beta for Claude Enterprise
The Claude Developer Platform has taken inference hooks out of beta for Claude Enterprise organizations, per Releasebot’s October 20 entry. The tracker gave few details beyond the headline, so check Anthropic’s release notes before building on it.
Amodei Answers Safety Critics in a 2,000-Word X Post
Responding to Encode’s Nathan Calvin, who accused Anthropic of weakening its safety commitments through $500 billion-plus in compute deals, growing reliance on Claude to build new models, and safety-policy changes, Dario Amodei listed nine commercially costly safety decisions. He said Anthropic was “the first, and still only, frontier company to restrict the sale of AI services to PRC-controlled companies.”
He acknowledged mistakes in how Anthropic communicated a draft federal AI safety bill and committed to propose options for regulating the pace of AI research by January 15, 2027, or explain why not. Fox News reports no response from Calvin or officials yet.
Safety Scrutiny Meets the IPO Runway
Two stories, one tension. Anthropic is defending its safety record in public while lining up what could be a record listing. The 5% fund is a clear gesture, but it is also a pre-IPO signal that the company wants its public-benefit story told on its terms.
Expect investors and Congress to ask the same question: does safety hold up under scale? Amodei’s January 15 promise gives critics a date to circle. Meanwhile the scorpion binder release shows the other side of the ledger, with careful caveats and open data.