Anthropic pledges 5% of its shares, pending approval, to a public benefit fund

A public benefit fund tied to the company's growth

On October 18, Anthropic announced that it is committing part of its own shares to the Anthropic Public Benefit Fund. Subject to stockholder approval, 5% of the shares will be committed at the outset. The board may add a further 2% for every 50% increase in the company's value above a set baseline. In Anthropic's words, the more the company grows, the larger the share that can be dedicated expressly to the public good.

Anthropic explains that most of its mission-related work already happens inside the company: AI safety and interpretability research, applied biomedical research, and discounts on Claude for nonprofits and scientific institutions. It says these activities have both public-benefit and for-profit value, which is why it keeps them in-house. The Fund is meant to add flexibility and scale, for example through large donations to traditional nonprofits or direct distributions to specific populations. Anthropic gives labor market impacts as one example. It says it has already committed $200 million to studying and addressing that issue, and that the pace of AI progress may call for bolder approaches. It chose a wholly-owned LLC rather than a 501(c)(3) to be able to adjust between forms of giving, and it states that none of the equity held by the LLC will be used for lobbying or political giving.

Anthropic also frames the move against executive pay. It says tech companies commonly offer escalating compensation awards to executives, and that it is directing this equity to the public benefit instead. It adds that the Fund builds on earlier commitments. The co-founders pledged to donate 80% of their equity at founding, and an equity donation match program lets employees pledge equity that Anthropic matches. Anthropic says approximately half of its employees take part. The text supplied to us is cut off partway through the sentence that combines these elements with the new announcement.

Sources: Investing in the public benefit — Anthropic

Research publications: alignment, quantum circuits and cryptography

Anthropic's Alignment Science blog published a post titled "Reducing Reward Hacking by Localizing Bad Behavior Instead of Penalizing It." The title sets out the contrast the authors draw between two ways of handling unwanted behavior, localizing it versus penalizing it. The summary we were given carries only the title, so we do not describe the method or the results.

Two other Anthropic items concern technical work done with Claude: "Designing more efficient quantum circuits with Claude" and "Claude reveals ECDSA curve origins." As with the alignment post, only the titles reached us. We therefore make no claim about what was found, how it was done, or how far the results go.

Sources: Reducing Reward Hacking by Localizing Bad Behavior Instead of Penalizing It — Anthropic Alignment Science blog · Designing more efficient quantum circuits with Claude — Anthropic · Claude reveals ECDSA curve origins — Anthropic

Anthropic on work, employment and expectations

Three further Anthropic publications share a theme of work and expectations: "The AI expectations gap," "How employment is changing at Anthropic," and "Working at the frontier shapes expectations for the future." The material provided contains only their titles, so we cannot say what they conclude.

The overlap with the day's main announcement is thematic only. The Public Benefit Fund text names labor market impacts as an area where Anthropic has committed $200 million, and these three titles also deal with work. We do not claim the publications are linked to the Fund.

Sources: The AI expectations gap — Anthropic · How employment is changing at Anthropic — Anthropic · Working at the frontier shapes expectations for the future — Anthropic

Claude Code 2.1.301 to 2.1.304: mod controls, auto mode and a closing date

Version 2.1.304 concentrates on mods. A mod's tool.check answer can now include decider: "person". In that case only the permission dialog or a host's callback can approve the ask. Where nobody can be asked, the request is refused with the mod's reason. The tool.check event also gains e.isReadOnly, set when the tool call is read-only. The release also adds a transcript line when a plugin that runs first skips one of a mod's hooks, naming the event and the plugin, so that a skipped hook is no longer silent.

Version 2.1.302 adds CLAUDE_CODE_AUTO_MODE_MODEL to choose the model used by the client-side auto mode classifier. It does not change the model for server-side auto mode. /config gains an Idle compaction row to turn idle compaction off or back on, and the notices shown after an idle compaction now explain how to turn it off. /design-sync and /design-login now display a notice that standalone Claude Design (claude.ai/design) closes by December 14, 2026.

Version 2.1.301 adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS. With it, unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG) ends a request at a usage limit that resets too far away, instead of waiting. The claude-api skill gains OpenTelemetry tracing guidance, including how to set up trace destinations with the ant CLI. Vim mode gains the % motion, which jumps to the matching bracket, alone or combined with d, c or y.

Sources: Claude Code 2.1.304 release notes — Anthropic (GitHub) · Claude Code 2.1.302 release notes — Anthropic (GitHub) · Claude Code 2.1.301 release notes — Anthropic (GitHub)