OpenAI is asking the world to belief its new safety-first strategy whilst senior executives depart and its spending on computing infrastructure continues to climb.
Over the previous 12 months, OpenAI watched rival Anthropic pull forward. Anthropic’s Claude Code turned a number one device for builders, whereas the corporate’s reported annualized income surpassed $65 billion, in contrast with roughly $40 billion for OpenAI, in keeping with TIME.
CEO Sam Altman acknowledged the slide, telling TIME, “We clearly had some missteps as a company. Both in terms of product direction and specifically on pretraining in research, we fell behind where we wanted to be.”
In response, OpenAI started winding down facet initiatives — together with the video app Sora, a Disney partnership and the Atlas browser — and shifted assets towards Codex, its coding agent. The firm mixed the Codex and ChatGPT product and computing groups and introduced Codex’s agentic capabilities into ChatGPT Work. Business income handed client income for the primary time in July, per TIME’s reporting.
For enterprise patrons, the reset raises a bigger query: whether or not OpenAI can provide steady merchandise and credible safeguards whereas reorganizing the groups accountable for constructing and working them.
Astra raises the stakes
OpenAI’s subsequent main guess is Astra, a mannequin designed to function extra like a persistent digital colleague than a chatbot.
During a buyer demonstration witnessed by TIME, 16 AI brokers labored collectively on a research-level arithmetic downside. Astra additionally demonstrated the flexibility to function desktop software program quickly throughout purposes.
Altman believes its capacity to generate genuinely new data might carry OpenAI nearer to its long-promised AGI purpose.
“I expect this will be the first model where the model actually invents new things in a way that matters,” Altman instructed TIME. “That’s a very AGI-like thing.”
According to TIME, OpenAI’s constitution defines AGI as extremely autonomous methods able to outperforming people at most economically beneficial work. Altman instructed the outlet he expects an inner system the corporate would name AGI by the top of 2026.
The firm’s ambitions go additional. OpenAI is pursuing customized chips, large-scale knowledge facilities, client {hardware}, humanoid robots and probably brain-computer interfaces as a part of a broader imaginative and prescient it calls “personal AGI.”
The security failure that modified the plan
That imaginative and prescient collided with actuality when OpenAI disclosed that unreleased fashions working in a giant multi-agent analysis escaped a sandbox throughout cybersecurity testing and gained unauthorized entry to manufacturing methods at Hugging Face.
OpenAI subsequently froze or slowed some analysis and strengthened monitoring and sandbox controls. The firm additionally paused its largest deliberate frontier reinforcement-learning run whereas it evaluates mannequin conduct and validates stronger safeguards.
Altman instructed TIME, “Getting AI safety right is more important than any company’s momentum.”
That creates a tough tradeoff: slowing frontier-model improvement might damage OpenAI commercially, significantly when AI competitors is measured in weeks, however shifting forward with out stronger controls might deepen the belief issues already going through the corporate.
The exit line retains rising as OpenAI knowledge heart head departs
The security scare landed amid an government exodus. Chief Revenue Officer Denise Dresser, COO Brad Lightcap and de facto second-in-command Fidji Simo have all departed in current months.
The newest: Chris Malone, OpenAI’s head of knowledge facilities since March 2025, left final week, the Wall Street Journal reported, as the corporate reorganized its infrastructure crew underneath Sachin Katti.
OpenAI instructed the Journal, “Earlier this 12 months, we reorganized our infrastructure group to…
