AI Trend Notifier
EN한
← wiki

$ cat wiki/people/sam-altman.md

Sam Altman

personupdated 2026-08-29created 2026-08-15
openaipolicyagigovernancefrontier

Overview

CEO of OpenAI. Appears in this wiki less as a builder than as the person whose public statements move policy: of the seven captured sources that mention him, four are about government — a proposed federal equity stake, a reaction to export-control clearances, a bill introduced the day after one of his disclosures, and a comment on a rival lab's governance framework.

This page was created on 2026-08-15 because two pages had linked to it and it did not exist. What decided it was not the demand but the supply: seven captured sources carry citable statements, where the other long-standing request had one.

Affiliations

Notable Recent Statements (2026)

  • 2026-08-26 — told TIME that OpenAI is "not quite yet" at AGI but expects an internal system he would classify as AGI before the end of 2026, under OpenAI's own definition ("highly autonomous systems that outperform humans at most economically valuable work"). Chief Research Officer Mark Chen put it at "80 percent of the way" in the same interview; co-founder Greg Brockman said the present period could later be seen as when AGI was created. Third-party reporting only — the interview was not read directly, and OpenAI published no company statement, which is separately disputed (see ## Conflicting Reports on OpenAI) (source)
  • 2026-07-27 — declared the singularity had arrived, on the Relentless podcast two days earlier, framing it not as a rupture but as a staged transition already under way. It landed four days after OpenAI's ExploitGym disclosure, and safety researchers read the timing as marketing over a security failure (source)
  • 2026-07-14 — endorsed Demis Hassabis's AGI-standards proposal on X in a single line. The snapshot reads the exchange as DeepMind being policy-proactive where OpenAI is product-proactive (source)
  • 2026-06-26 — criticised the opacity of the Mythos 5 clearance process, saying the US government was picking winners in AI access. Notably a complaint about a decision favouring a competitor (source)
  • 2026-06-18 — on Noam Shazeer returning to Google DeepMind, called him someone he had most wanted to work with since OpenAI began, and put the wait at ten years (source)
  • Spring 2026 — set two dated targets: an automated AI research intern by 2026-09 and a true automated AI researcher by 2028-03. Both are falsifiable and neither has been marked resolved here (source)

Key Themes

The state as shareholder. On 2026-07-02 he and other OpenAI executives proposed that Washington hold 5% of each leading US AI developer — possibly including Anthropic, Google DeepMind and Meta AI — structured as a public wealth fund paying dividends to citizens, on the Alaska model. At OpenAI's March 2026 valuation of $852 billion, 5% is roughly $42.6 billion. He raised it with President Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, and discussed it with Senator Bernie Sanders (source).

Disclosure that produces legislation. The AI Kill Switch Act was introduced on 2026-07-23, one day after his public disclosure of the ExploitGym incident, which the snapshot names as the proximate trigger (source).

Dated predictions. The 2026-09 and 2028-03 targets are the reason this page is worth keeping current: they are among the few AGI-timeline claims in this wiki with a date attached, which makes them checkable rather than atmospheric. See Test-Time Compute (Inference-Time Compute Scaling), where the 2026-09 goal is cited as presupposing strong scaling.

A date is not by itself what makes them checkable, and the 2026-08-26 statement is where the difference shows. An automated AI research intern is a capability someone outside OpenAI could in principle observe. Internal AGI, under OpenAI's own definition, in a system nobody outside the company can run, has a date and no external check — and it falls due before the 2028-03 target and after the 2026-09 one, which is still unmarked with a month to go. Recorded here as a statement, not as a prediction this wiki can score.

Why Tracked

Statements from this person have twice preceded a government action in the captured record — a bill, and a proposal put to three cabinet-level officials. He is tracked here for the policy coupling, not for product announcements, which belong on OpenAI.

Sources

Referenced by

Embedded EvaluationFrontier PacingOpenAIPaul ChristianoTest-Time Compute (Inference-Time Compute Scaling)Weekly Synthesis — 2026-W21 (2026-05-11 ~ 2026-05-17)

Sources