OpenAI's GPT-5.6 model shows autonomous self‑training as US denies approval

GPT-5.6 autonomous self‑training recursive self‑improvement White House denial OpenAI

OpenAI unveiled its latest AI family – GPT‑5.6 Sol, Terra and Luna – and demonstrated that the Sol variant can autonomously post‑train the smaller Luna model. Using a "fairly under‑specified prompt" on the Codex platform, Sol selected training configurations, allocated GPUs, launched the script and verified the run, improving Luna without human intervention. In internal tests Sol scored 16.2 points higher than GPT‑5.5 on OpenAI’s Recursive Self‑Improvement benchmark, and researchers reported a doubling of daily token output per active researcher.

The rollout sparked political controversy after reports claimed the Trump administration had given OpenAI a "green light" for release. The White House rejected the claim, with a spokesperson stating, "The Trump administration did NOT give OpenAI a 'green light,' approval, or clearance to release its models." The statement emphasized that no mandatory licensing is required under the June 2 executive order on voluntary AI safety testing.

Sources

OpenAI's GPT-5. 6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt" [the-decoder.com]

11 days ago

White House Says It Never Approved OpenAI's GPT-5. 6 Public Release [www.timesnownews.com]

12 days ago