OpenAI published a plan on June 8, 2026, the day before its confidential S-1 filing became public, and the timing was not accidental. CEO Sam Altman and Chief Scientist Jakub Pachocki co-authored the piece, framing what they call the company’s “third phase” around three explicit goals: build an automated AI researcher by March 2028, accelerate broad economic gains, and give every person on Earth a personal AGI. Read as a mission statement, it is idealistic. Read as IPO narrative scaffolding, it is precise.
The automated-researcher target is the most specific claim in the document. Altman and Pachocki write, per openai.com, that they expect a significant fraction of OpenAI’s own research to be conducted by AI systems working in tandem with human researchers by March 2028. That is twenty-one months from publication. The claim is specific enough to function as a milestone for investors, and vague enough (a “significant fraction” is undefined) to survive if reality falls short.
The safety argument the essay advances is worth reading twice. The core claim is that concentrating AGI in few hands is more dangerous than distributing it widely. This is a logically coherent position. It is also the position most favorable to a company that is about to raise public capital and needs to justify aggressive deployment rather than caution. The essay acknowledges no tension between those two facts.
Compare that to Anthropic’s essay from the same week, in which the company argued for an optional global pause mechanism tied to recursive self-improvement. Anthropic’s framing treats capability acceleration as a contingency requiring a circuit breaker. OpenAI’s framing treats acceleration as the safety strategy itself. Two companies valued, collectively, north of a trillion dollars took opposite policy positions within days of each other. Both publish these positions to the same institutional investor audience. The positions are not incidentally different; they reflect genuinely different bets on what governance structure protects the world and, not coincidentally, which business model it rewards.
The Personal AGI framing deserves separate attention. Altman and Pachocki use empowerment language rather than automation language, describing a future where each person directs an AGI toward whatever they choose. That framing is the consumer-distribution play that justifies the company’s product investment cycle: the ChatGPT overhaul, the Codex toolchain, the Dreaming memory architecture announced last week. If every person on Earth is the addressable market, the consumer revenue multiple is very large. IPO analysts will note this.
The “third phase” structure is doing rhetorical work as well. Phase one was proving AI was possible. Phase two was scaling it. Phase three is, per the essay, making it abundant and affordable. That arc positions every prior capability milestone as infrastructure for the monetization phase. That is a growth-equity narrative, not a safety narrative, and it is the one institutional investors will act on.
What the essay does not include: independent validation of the March 2028 researcher target, any mechanism for measuring whether gains are “broadly shared,” or any definition of the threshold at which distribution becomes dangerous rather than protective. The release announcement does not include an external advisory opinion or a governance body that could hold the timeline accountable.
The Anthropic contrast is the clearest way to read what OpenAI chose. Anthropic accepted a constraint on its own acceleration in the name of systemic safety. OpenAI declined the same framing and called acceleration itself the safety mechanism. Both labs are raising at historically high valuations. One of these positions will look correct in hindsight. The market will price that disagreement in the next six months.
Teams evaluating which frontier lab to build on through a multi-year contract should note that this policy divergence is now public, explicit, and attached to capital-raising events. The lab you choose is also a governance bet.
Sam Altman and Jakub Pachocki via OpenAI (openai.com), 2026-06-08.
