AI Industry Daily News

OpenAI Limits GPT-5.6 Sol Release at Trump Administration's Request

OpenAI announced a limited preview of GPT-5.6 Sol. CEO Sam Altman reportedly told staff the staggered approach follows discussions with the Office of the National Cyber Director and the Office of Science and Technology Policy, citing cybersecurity concerns. The Trump administration asked OpenAI to release the model gradually rather than all at once, with the government approving customer access individually before a planned general release in a few weeks.

The move reflects rising national-security and cyber-threat concerns over rapidly improving frontier AI, and sets a precedent for pre-release government review. It was reported by Bloomberg, The Guardian and others. In OpenAI's internal CTF tests, all models exceeded the 'High' cybersecurity threshold, with Sol reaching 96.7%.

The GPT-5.6 series comprises three models including the flagship Sol. On Terminal-Bench 2.1, Sol Ultra scored 91.9% and Sol 88.8%, and Sol reached 50.9% on Agent's Last Exam in code mode.

Anthropic Re-enables Claude Mythos 5 and Fable 5 Access for US-Approved Critical Infrastructure

From June 12, Anthropic announced it can re-enable access to its top-tier cybersecurity models Claude Mythos 5 and Fable 5 for US critical-infrastructure organizations approved by the government, in close coordination with federal authorities.

Rather than opening the powerful models to all users, Anthropic adopted a cautious distribution approach gated by government coordination and limited use cases. Observers noted this access-control model could become standard for highly capable models.

Similar government-conditioned, staged model availability is spreading across the industry. The Information reported OpenAI is staging GPT-5.6's release at the government's request, while Anthropic at times paused model availability amid AI-related export controls.

OpenAI Said to Delay IPO to 2027 as Altman Rejects Sub-$1T Valuation

OpenAI is reported to have delayed its IPO until 2027, with CEO Sam Altman rejecting a valuation below $1 trillion as a 'nonstarter.' SoftBank shares dropped sharply on related news.

Separately, The Information noted that after SpaceX completed its record IPO just 74 days from its confidential S-1 filing, OpenAI and Anthropic are increasingly likely to file public prospectuses within weeks. Anthropic filed a confidential S-1 around June 1 with a reported pre-valuation of about $965 billion, and OpenAI is said to have filed around June 8-9 with an estimated valuation of $850 billion to $1 trillion.

SoftBank's Masayoshi Son Questions Musk's Orbital AI Data Center Plan

At a SoftBank mobile-unit shareholder meeting around June 23-24, Son responded to a shareholder question by asking what point there is in building AI data centers in space, signaling a preference for ground-based construction. He argued power makes up only about 7% of data center operating costs, which cannot justify the complexity of launch, maintenance and latency, and that AI's outcome will be decided in 'the next few years.'

TechCrunch on June 27 expanded on the comments, noting space data centers are not an immediate solution and that satellites needing replacement every few years feed directly into SpaceX's launch business profits.

Meanwhile, SpaceX unveiled the design of its first orbital AI data center satellite, AI1, built on Starlink V3 technology and using solar power with radiative cooling into vacuum. SpaceX filed with the FCC in January 2026 for a constellation of up to one million satellites, with similar efforts from Google's 'Suncatcher' and Starcloud.