Altman to Brief the White House Next Week on Next-Gen Models｜Chatter Says GPT-6 Has Been in Internal Testing for 2.5 Months
TL;DR
Sam Altman will brief the Trump administration and Congress on OpenAI's next-gen models next week — as social chatter claims GPT-6 has been in internal testing for 2.5 months.
Sam Altman will brief the Trump administration and Congress next week on OpenAI's next generation of AI models. Chris Lehane, OpenAI's head of global public affairs, confirmed this at a July 21 Washington briefing.
Timing matters — the US government is finalizing a safety review framework for frontier AI systems, expected to land in the coming weeks. Altman's visit lands right as the framework closes. The read — OpenAI wants to get its next-gen model's capability envelope in front of policymakers before the framework hardens, avoiding a later block.
Lehane's specific language — "this family of models has some very interesting capabilities, particularly around work and extending work". That phrasing is rare in OpenAI public copy — "extending work" usually implies models that can sustain multi-day autonomous tasks, approaching the "sustained autonomy" bar in AGI definitions.
Social chatter runs louder — X user @daniel_mac8 posted that GPT-6 has achieved AGI, citing a one-shot counterexample to the Jacobian conjecture (an open algebraic-geometry problem for nearly 90 years). The post cites a prediction — GPT-6 has been in internal testing at OpenAI for about 2.5 months, with public release possibly earlier than expected.
The timing lines up with Altman's briefing — 2.5 months of internal testing means capability has stabilized, and now the play is "show the government, build regulator trust, prep the launch". Same script as pre-GPT-5.
Comparison — Google DeepMind just confirmed Gemini 4 has entered pretraining (see #406), Anthropic Opus 4.8 is shipping, OpenAI GPT-5.6 Sol holds the flagship board — if GPT-6 lands before Q4, Google's Gemini 4 cadence is thrown.
via Bloomberg / Dan McAteer on X
Timing matters — the US government is finalizing a safety review framework for frontier AI systems, expected to land in the coming weeks. Altman's visit lands right as the framework closes. The read — OpenAI wants to get its next-gen model's capability envelope in front of policymakers before the framework hardens, avoiding a later block.
Lehane's specific language — "this family of models has some very interesting capabilities, particularly around work and extending work". That phrasing is rare in OpenAI public copy — "extending work" usually implies models that can sustain multi-day autonomous tasks, approaching the "sustained autonomy" bar in AGI definitions.
Social chatter runs louder — X user @daniel_mac8 posted that GPT-6 has achieved AGI, citing a one-shot counterexample to the Jacobian conjecture (an open algebraic-geometry problem for nearly 90 years). The post cites a prediction — GPT-6 has been in internal testing at OpenAI for about 2.5 months, with public release possibly earlier than expected.
The timing lines up with Altman's briefing — 2.5 months of internal testing means capability has stabilized, and now the play is "show the government, build regulator trust, prep the launch". Same script as pre-GPT-5.
Comparison — Google DeepMind just confirmed Gemini 4 has entered pretraining (see #406), Anthropic Opus 4.8 is shipping, OpenAI GPT-5.6 Sol holds the flagship board — if GPT-6 lands before Q4, Google's Gemini 4 cadence is thrown.
via Bloomberg / Dan McAteer on X
