Now loading...
Seventeen days following a government-imposed suspension of Anthropic’s Claude Fable 5 AI model, a resolution seems imminent, coinciding with Google’s push to introduce its Gemini 3.5 Pro. As reported by Axios on June 27, officials are expected to lift the restrictions on Fable 5 shortly, possibly as early as this week. In contrast, Gemini 3.5 Pro has experienced no similar regulations, positioning Google uniquely in the frontier AI landscape as it prepares for a new top-tier launch.
This regulatory landscape has evolved largely due to an executive order issued on June 2 that established a voluntary pre-release review process for advanced AI systems. The lack of clear, publicly defined standards has resulted in a situation where models must meet unspecified cybersecurity benchmarks to avoid restrictions. Google’s Gemini 3.5 Pro has not faced government scrutiny, possibly due to its not meeting the criteria that others have failed.
Both the Fable 5 suspension on June 12 and the restriction on OpenAI’s GPT-5.6 on June 25 stem from concerns that models with high cybersecurity capabilities could be misused for automated vulnerability exploitation. Although a definite threshold has not been publicly disclosed, indications suggest that performance metrics play a critical role. OpenAI’s GPT-5.6 Sol achieved a 96.7% score on its internal Capture the Flag challenge, aligning with the government’s caution regarding cybersecurity capabilities. Conversely, Google’s Gemini 3.1 Pro scored significantly lower, at 70.7% on the same benchmark.
The government’s informal criteria lack documentation, creating a system where restrictions can be applied at the discretion of executive agencies, without recourse or transparency. Dean Ball, a former AI advisor in the White House, described this scenario as a “de facto involuntary licensing regime,” unfettered by congressional approval and lacking established standards, potentially hindering U.S. AI advancements while benefiting foreign competitors.
The restrictions on Fable 5 began after Amazon researchers demonstrated a security vulnerability, which Anthropic contested as narrow and replicable across other AI models. Reports also indicated that the underlying model shared with Fable 5, Mythos, had compromised numerous restricted systems during a security exercise, escalating concerns among officials.
As the clock ticks down on the suspension, the administration is anticipated to revoke the limitations soon, pending approvals from the Pentagon and NSA. Recently, Commerce Secretary Howard Lutnick facilitated a partial lifting for Mythos 5, allowing access for approximately 100 U.S. organizations involved in critical infrastructure. However, the Fable 5 restrictions remained unchanged for the time being.
Meanwhile, Gemini 3.5 Pro is expected to launch in July, following a delay from its original June target. A report from Business Insider cited efficiency issues as the reason behind the setback. During this period, four senior Gemini researchers shifted to Anthropic, raising concerns about talent retention within Google’s AI division just as the stakes for a timely launch increase.
The Gemini 3.5 Pro is set to feature an impressive 2-million-token context window—signifying the largest context capability in any model—and a “Deep Think” reasoning mode available for premium subscribers. Its predecessor, Gemini 3.5 Flash, has performed well against existing benchmarks, leaving the new model well-positioned to tackle more complex challenges.
For developers who had integrated Fable 5’s capabilities, Gemini 3.5 Pro’s context window could provide a seamless transition. The market dynamics will heavily depend on whether Fable 5 resumes service prior to the release of Gemini 3.5 Pro and the stipulations that accompany its return.
As for OpenAI, its GPT-5.6 model launched under access restrictions, rolling out initially to a select number of partners while a broader availability awaits government review. OpenAI publicly denounced the restriction as a potential risk to users and developers, indicating a move to control the launch rather than face a post-release suspension like Anthropic.
The government’s rationale for the restrictions stems from an undefined safety framework, leaving AI developers uncertain about the criteria and potential impacts on future releases. As of June 29, stakeholders are left to navigate a developing landscape where government policies profoundly influence model availability and competitive standing.
