OpenAI Seeks AI Limits; 2 Rivals Dropped a Stop in 2026

OpenAI asked rival labs on Sunday to agree common limits they would not cross until AI monitoring catches up. We read the latest public safety frameworks we found from four rivals: Anthropic, Google DeepMind, Meta and xAI.

By our count, 2 of the 4 softened a stop or pause commitment in 2026. Meta’s own change log shows its top risk threshold went from “Stop” to “Develop with Mitigations” on 7 April.

Anthropic dropped its pause pledge in February, Semafor reported. All four companies signed the White House AI accord in September. We track Anthropic on our Anthropic stock page.

Business Pill · The Cost of Checking

An evergreen Business Engineer explainer. It teaches the general idea only and does not discuss this case.

The key insight: OpenAI proposed common limits on 22 November 2026; by our count, 2 of 4 rivals whose latest frameworks we read (Meta and Anthropic) softened a stop or pause commitment earlier in 2026, and only xAI’s framework sets numeric thresholds.

What OpenAI Asked For

Chief scientist Jakub Pachocki proposed “common limits, based on perplexity and other measurements, that developers are committed to not crossing until our ability to understand and monitor AI behavior achieves a certain concrete milestone.”

He wrote that he hopes it “leads the way to a common framework for setting successive limits and pacing further development.”

What four of OpenAI’s rivals have committed to, from their latest public frontier-safety documents we fo
What four of OpenAI’s rivals have committed to, from their latest public frontier-safety documents we found. Meta and Anthropic softened a stop or pause commitment in 2026.

Meta: From Stop to Mitigations

Meta renamed its Frontier AI Framework the Advanced AI Scaling Framework in version 2. Its change log, dated 7 April 2026, says the critical threshold measure changed from “Stop” to “Develop with Mitigations.”

The log adds: “All thresholds now permit proceeding with sufficient mitigations validated to reduce risk to moderate or lower.” Meta also replaced “uniquely enable” with “substantially contribute to” as its test for high or critical risk.

Anthropic: Goals, Not Hard Commitments

Anthropic published version 3.0 of its Responsible Scaling Policy on 24 February. “Anthropic dropped its pledge to pause development if the capability of its AI models far outstripped safety research,” Semafor reported the next day.

Anthropic wrote that “we found pre-set capability levels to be far more ambiguous than we anticipated.” Its new roadmap goals are, in its words, “rather than being hard commitments, these are public goals that we will openly grade our progress towards.”

The policy also sets out what Anthropic calls “an ambitious capabilities-to-mitigations map” for the whole industry. As we read it, that is close in spirit to the cross-lab rules OpenAI now wants.

Timeline of 2026: 24 Feb Anthropic RSP v3.0 (pause pledge dropped per Semafor), 7 Apr Meta framework v2 (Stop to Develop with Mitigations), 17 Apr Google DeepMind update, 29 Sep White House accord, 22 Nov OpenAI proposes common limits
In 2026, Anthropic and Meta softened stop or pause commitments before OpenAI proposed common limits on 22 November. Sources: company documents; Semafor.

Google DeepMind and xAI

Google DeepMind’s Frontier Safety Framework is built around Critical Capability Levels. Its update post says: “we conduct safety case reviews prior to external launches when relevant CCLs are reached.”

xAI’s Frontier AI Framework, last updated 30 December 2025, is the only one of the four where we found numbers. Its deployment criterion is an answer rate below 1 in 20 on restricted biology and chemistry queries, and a dishonesty rate below 1 in 2 on the MASK benchmark.

The Structural Read

As we read it, OpenAI’s proposal runs against the 2026 trend among its rivals, which moved from fixed stops toward proceeding with mitigations.

As we read it, Anthropic’s industry-wide capabilities-to-mitigations map is the closest existing text to what OpenAI is asking for.

Meta, Advanced AI Scaling Framework v2 change log, 7 Apr 2026

“All thresholds now permit proceeding with sufficient mitigations validated to reduce risk to moderate or lower.”

Three Implications

FOR OPENAI Pachocki asked developers to commit to common limits until AI monitoring reaches a concrete milestone.

FOR META AND ANTHROPIC Meta’s critical threshold now reads ‘Develop with Mitigations’; Anthropic’s roadmap goals are public goals rather than hard commitments, per their documents.

FOR POLICYMAKERS xAI’s framework is the only one of the four with numeric deployment thresholds we found.

What Is Not Established

We read Meta’s and xAI’s full framework texts and the Anthropic and Google DeepMind update posts, not every page of their full policies. For Anthropic we cite the version 3.0 announcement, not later revisions.

By our word search, “perplexity” appears in none of the four documents we read. By publication, we had not seen any of the four respond on record to OpenAI’s proposal. Our earlier comparison of OpenAI’s model-stop figure with rival labs is here.

This Week In The AI Supercycle

A World Without an Order, and an AI Central Bank Without a Successor

This week’s issue: the week AI memory stayed largely sold out while the people holding the shares paid: why the AI Supercycle is not a dot-com replay, read through geopolitics, macro and technology, with the gap between building and financing the widest we’ve recorded.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of who builds, finances and captures value in AI.

Read the Map of AI →

The Bottom Line

OpenAI is asking for shared limits in a year when two of its biggest rivals loosened their own. Meta and Anthropic now describe proceeding with mitigations or goals rather than a hard stop. Whether they sign up to a common line is the next thing to watch.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Quotes and thresholds are from each company’s own document: Meta’s Advanced AI Scaling Framework v2 change log (7 April 2026), Anthropic’s RSP v3.0 announcement (24 February 2026), Google DeepMind’s Frontier Safety Framework update post and xAI’s Frontier AI Framework (30 December 2025). The Anthropic pause change is as reported by Semafor. The count of 2 of 4 and the word search are ours.

Sources: OpenAI (Jakub Pachocki), ‘Perplexity and Intelligence’, 22 Nov 2026 · Meta, Advanced AI Scaling Framework v2 (change log 7 Apr 2026) · Anthropic, Responsible Scaling Policy v3.0 announcement, 24 Feb 2026 · Semafor, ‘Anthropic eases AI safety restrictions to avoid slowing development’, 25 Feb 2026 · Google DeepMind, ‘Strengthening our Frontier Safety Framework’ (updated 17 Apr 2026) · xAI, Frontier Artificial Intelligence Framework (last updated 30 Dec 2025)

Scroll to Top
FourWeekMBA