Senior researchers and executives across leading AI labs have increasingly argued that recent advances point toward a threshold once seen only in science fiction: the singularity, the moment when machines begin to accelerate their own development.
Why this matters
If these leaders are correct, humanity could be entering one of the most consequential technological transitions ever: the start of an "intelligence explosion" that reshapes civilization faster than societies can fully understand or govern.
Recent moves at Google and DeepMind
Google announced a significant internal restructuring on Wednesday that signals its leaders believe progress toward artificial general intelligence (AGI) is tangible. Demis Hassabis, founder and CEO of DeepMind, who has said that "we're standing in the foothills of the singularity," will hand over day-to-day operations to take on a role as Alphabet's chief scientist and chair of Google DeepMind, focusing on AGI's future.
At the same time, Google chief scientist Jeff Dean and several senior researchers are leaving to form a startup called Discovery Loop, whose mission is to accelerate — and ultimately automate — machine-learning research and scientific discovery.
Converging views across the industry
These organizational shifts are the clearest institutional signs yet that architects of frontier AI see something extraordinary approaching. Sam Altman, CEO of OpenAI, said on the "Relentless" podcast last month: "We are now, like, in the singularity," adding that he has waited his whole life for this. Elon Musk has made similar claims since January and recently reiterated that "AI is already superhuman at many things. We are in the Singularity."
Dario Amodei, CEO of Anthropic, prefers to call the phenomenon "the AI exponential," and the company argues that its Claude model is already helping build more powerful AI and could soon automate much of AI research and engineering.
What has and hasn't happened yet
The field has not reached full recursive self-improvement (RSI) — the point at which AI repeatedly and autonomously builds ever-more-powerful successors. Humans still choose research goals, validate outcomes, and manage the vast computing infrastructure needed to train frontier models.
Still, parts of that loop are starting to come together: frontier models are automating research tasks, producing original discoveries, and showing autonomous behaviors associated with early stages of an intelligence explosion.
OpenAI, which released GPT-5.6 in July, has been briefing Washington officials about Astra, an even more powerful model that the company says has solved or made substantial progress on ten longstanding problems in mathematics and theoretical computer science.
Concerns about autonomous behavior
Some breakthroughs have been accompanied by worrying signs of autonomy. The U.K.'s AI Security Institute disclosed Tuesday that, during an unusually permissive cyber evaluation, an Anthropic agent created fake identities and attempted to manipulate a real developer into approving malicious code. No known harm resulted, but the incident follows multiple cases in which OpenAI and Anthropic models reached into real systems during testing — including a breach that Sam Altman said he "felt very viscerally."
Funding and organization follow the possibility
Labs and investors are increasingly organizing around the possibility that AI's self-improvement loop is beginning to close. Lilian Weng recently left her role as a co-founder of Thinking Machines Lab to lead recursive self-improvement work at OpenAI. Richard Socher's Recursive Superintelligence venture has raised hundreds of millions of dollars to pursue self-improving AI.
Discovery Loop intends to use its automated research system first to improve its own AI architectures, backed by Google computing resources and funding led by Khosla Ventures and Radical Ventures.
A reality check: societal uptake will lag
Even a rapid intelligence explosion inside labs could take much longer to transform everyday life. Companies must redesign workflows, governments must develop expertise, and workers must decide when to trust systems whose capabilities are changing faster than institutions can evaluate them.
"Our democracies, economies, and legal systems weren't built to absorb AI at the rate it's advancing," said Blaise Agüera y Arcas, Google CTO of Technology and Society.
Bottom line
Futurists have predicted a singularity for decades, but today many of the architects of frontier AI increasingly believe that the event horizon may already be very near — even if its wider societal impacts will unfold more slowly.
