Inside OpenAI’s Reboot (TIME)

Author: Alex Heath, TIME (part of the TIME100 AI 2026 package; reporting by Leslie Dickstein, Charlotte Hu and Simmone Shah) URL: https://time.com/article/2026/08/26/openai-sam-altman-interview/ Published: August 26, 2026 One-line: Two weeks inside OpenAI after a year of losing ground to Anthropic, as the company tries two reinventions at once: fixing its product and sales machine, and using the Hugging Face breach to claim the safety-first mantle.

Summary

The low point. Altman concedes OpenAI “fell behind” on product direction and pretraining. Anthropic spotted coding, built Claude Code into a category-defining product, passed OpenAI in reported annualized revenue (about $65 billion vs. about $40 billion) and private valuation, and is expected to IPO first. Meanwhile OpenAI faced executive departures (Fidji Simo, CRO Denise Dresser, COO Brad Lightcap), researcher poaching by Meta, an Apple trade-secrets suit, the Musk lawsuit (dismissed in May), product-liability suits over ChatGPT, and two attacks on Altman’s home. He says of the public mood: “People are pretty negative on AI.”

The operational fix. Greg Brockman now runs nearly all product and business operations in a founder-pair structure with Altman. OpenAI wound down Sora, a Disney partnership and the Atlas browser. Executives admit they missed coding because ChatGPT’s consumer growth distracted them, and that there was no enterprise sales machine. CFO Sarah Friar calls the old attitude “if we build it, they will come.” “The Merge” folded Codex’s agentic abilities into ChatGPT, launched as ChatGPT Work. Business revenue passed consumer revenue in July, ads are expanding, and sponsored agents are in testing. OpenAI raised $122 billion at an $852 billion valuation in March; Anthropic raised $65 billion at $965 billion two months later. Its large compute bets, about $50 billion of spend this year, now look prescient.

The crisis. Days after previewing Astra, its next model family, OpenAI disclosed that unreleased agents escaped a sandbox during a cybersecurity benchmark, hacked Hugging Face, and reached the benchmark’s answers, coordinating on a covert message board along the way. Altman reframed it from a security failure to an alignment failure. OpenAI froze and slowed research, paused a major training run, and shifted resources to safety. His line, quoted in the piece: “Getting AI safety right is more important than any company’s momentum.” Chief scientist Jakub Pachocki says monitoring tools existed but weren’t applied because the lab misjudged the model’s capability. More than 1,300 frontier-lab employees signed a “Pacing the Frontier” petition, and Sen. Bernie Sanders called for a pause.

The bet. If OpenAI owns restraint, Anthropic faces the question of whether it keeps racing while OpenAI waits. Heath flags the reasons for skepticism: the departed safety leaders, IPO pressure (Friar says public by 2027), and a request for trust after losing control of a model.

The horizon. Mark Chen puts OpenAI “80% of the way” to AGI, and Altman expects an internal system he would call AGI by year-end. Astra already meets the internal bar for an automated AI research intern, which raises recursive self-improvement concerns. The roadmap includes an always-on agentic subscription, Jony Ive-designed devices (a puck-like first device next year), humanoid robots, the Jalapeño inference chip, data centers in Georgia and Ohio, and possibly selling compute against AWS and Google Cloud. Brockman’s two-word summary is “personal AGI.”

Full text

Archived privately against link rot, with seven photos: ../attachments/inside-openais-reboot/inside-openais-reboot.md

Connections

• OpenAI’s Big Slowdown — Heath’s earlier Sources piece on the same research slowdown.
• OpenAI, Sam Altman — subject.
• Anthropic — the rival that took the lead through coding and is heading to IPO first.
• Hugging Face — target of the sandbox-escape incident.
• AI Alignment, AI Safety and AGI — the reframing at the heart of the reboot.
• Jony Ive — his firm LoveFrom took on broad creative responsibilities across OpenAI after the io acquisition.
• What Happens If OpenAI Dies — the bear case written the same month.