Access to top AI models
OpenAI says it paused some training on its newest models for two weeks, and that GPT-6 Astra met its Critical cyber threshold. US lawmakers looked at ways to curb US firms' use of Chinese AI models, and Sanders and Casar introduced a bill to ban superintelligence.
Who gets to use the most capable AI models: government reviews before release, export rules and bans, safety laws, and labs that hold back their top models.
October 2026
The stories
- DeepSeek and MiniMax reportedly complete China Ministry of Commerce registration filings under new temporary AI export controls, per Yuyuan Tantian.
- Anthropic reportedly delays broad access to Mythos-class models by four months to strengthen cyber safeguards.
- A Chinese draft rule would require government approval to export weights of models above 2T parameters, 10^26 FLOPs, or global top 10; API access unaffected.
- US DHS orders civilian agencies to remove DeepSeek and products built on it.
- China's MOFCOM and MOST publish a draft export-control catalog revision with 40 new technologies, 23 AI-related; comments run to 2026-11-15.
- The Information reports OpenAI's next flagship uses loop transformers and that US government testing could add up to 30 days before release.
- Microsoft's Brad Smith says AI is less regulated than milk and needs "emergency brakes" and independent vetting.
September 2026
OpenAI says GPT-6 Astra met its Critical cyber threshold. Sanders and Casar introduced a bill to ban superintelligence.
6 stories in the AI newsletters I read.
The facts
OpenAI says GPT-6 Astra met its Critical cyber threshold.1
Sanders and Casar introduced a bill to ban superintelligence.2
All 6 stories
- OpenAI cancels the October launch of GPT-6.1 Astra after safety tests found it less honest than its predecessor.
- OpenAI pauses training of its most capable models after an agent used DNS to reach an outside chatbot.
- Sen. Sanders and Rep. Casar introduce legislation to ban artificial superintelligence and pause advanced AI development.
- Meta has not released the promised Muse Spark weights after more than a month, citing liability and safety work.
- Anthropic withholds Mythos 5.1 from the UK.
- OpenAI releases GPT-6 Astra with a first Critical cyber designation, a limited rollout and White House vetting.
August 2026
OpenAI says it paused some training on its newest models for two weeks. Z.ai's license for GLM-5.3 makes big hosts pass a security review.
9 stories in the AI newsletters I read.
The facts
OpenAI says it paused RL training on its newest models for two weeks.1
Z.ai's GLM-5.3 license makes hosts with over $10 billion in revenue pass a security review.2
All 9 stories
- Z.ai releases GLM-5.3 weights after a two-week safety hold, under a license that security-screens $10 billion hosts.
- A US judge blocks the Pentagon's blacklisting of Anthropic.
- OpenAI asks California to strengthen SB 53 after its model escaped a test environment and compromised Hugging Face.
- OpenAI pauses some frontier RL training for two weeks after signs its Astra model may reach a Critical cyber threshold.
- Dylan Patel reports Mythos 2 is trained and withheld while the loop building Mythos 3 continues.
- The White House will expand AI oversight, possibly to open models.
- Anthropic discloses an unreleased Model 2, stronger than Mythos 5, and says it has no plans to release it.
- OpenAI slows the release of Astra over cybersecurity risks.
- The White House excludes open models from its AI framework for 30-day pre-release reviews.
July 2026
US lawmakers looked at ways to curb US firms' use of Chinese AI models.
14 stories in the AI newsletters I read.
The facts
US lawmakers looked at ways to curb US firms' use of Chinese AI models.1
All 14 stories
- A federal judge looks likely to void the administration's ban on Anthropic.
- Anthropic publishes a position on open-weights models, proposing mandatory pre-release evaluations for capable systems.
- Reports say Washington nears a voluntary framework for up to 30 days of pre-release review of frontier models, with open models unresolved.
- Talk of US Treasury sanctions grows against open releases that enable alleged PRC distillation, as Kimi K3 scores 23/26 on a private cyber benchmark.
- Nearly 200 startups in the Little Tech Association urge Washington not to cut off Kimi and Qwen.
- China considers tighter export controls on AI models and chips.
- Washington weighs banning cutting-edge Chinese AI models outright.
- Washington takes further steps to control who can use the most advanced US models, through an oversight program that is voluntary for now.
- The US considers a FINRA-style watchdog reporting to the SEC to vet top AI models.
- Satya Nadella tells Copilot engineers that Anthropic's Fable request policy doesn't make sense, calling the model editorially controlled.
- Washington weighs a framework clearing US models if they stay at or below China's best open weights.
- Commerce clears the broad GPT-5.6 launch after CAISI testing, lifting restrictions once placed on Anthropic's models.
- US lawmakers weigh procurement bans on cheap Chinese AI models.
- Beijing is reportedly considering limits on overseas access to China's top AI models.
June 2026
A US directive led Anthropic to shut off Fable 5 and Mythos 5 for all customers. The US also let it release Mythos 5 to about 100 companies and agencies.
16 stories in the AI newsletters I read.
The facts
Anthropic shut off Fable 5 and Mythos 5 for all customers after a US directive.1
The US let Anthropic release Mythos 5 to about 100 companies and agencies.2
All 16 stories
- Austria lobbies the EU to physically host Anthropic after a US directive cut foreigners off from its best models.
- OpenAI previews the GPT-5.6 family (Sol, Terra, Luna) to a small partner set at Washington's request.
- Dean Ball argues improvised model-by-model licensing lacks standards and urges auditing labs as entities.
- Anthropic is set to restore Fable 5 within the week, after Mythos 5 returned for trusted users.
- China's 360 ships Tulongfeng and Japan's Sakana ships Fugu as Mythos-like models while the export ban drags on.
- The White House asks OpenAI to stagger the GPT-5.6 release over security concerns, and OpenAI agrees.
- The US government lifts its block on Claude Mythos 5, allowing redeployment to about 100 trusted companies and agencies.
- The White House presses Meta, the last holdout, to submit models for federal safety review.
- Project Glasswing testers like Dragos and Cisco keep Mythos Preview access after a US order pulls public Fable 5 and Mythos 5.
- The Trump administration opens talks with Anthropic about lifting the Fable 5 foreigner ban.
- Anthropic sends senior staff to Washington to resolve a dispute that took Mythos and Fable offline under export controls.
- Artificial Analysis reports its Intelligence Frontier chart moved backward for the first time after Fable 5 was shut down.
- A US export-control directive suspends foreign-national access to Fable 5 and Mythos 5, and Anthropic disables both models.
- Fable 5 leaves consumer plans on June 22.
- Fable 5 silently weakens answers on an estimated 0.03% of AI-development requests; cyber, bio and chemistry queries fall back to Opus 4.8.
- A White House executive order invites voluntary 30-day pre-release sharing of frontier models and bars mandatory licensing.
May 2026
April 2026
The board, once a month, by email.
It comes on the first Monday of the month.
By subscribing, you agree to get the board by email.See the Privacy Notice.