Key Notes
- Anthropic safeguards researcher Tao Lin calls for AI research to slow to roughly one quarter of its current pace to allow more safety work.
- He forecasts broadly superhuman AI around late 2027 while expecting job displacement to take longer because robots and other infrastructure must be built.
- His risk estimates and timelines are personal judgments, and the interview does not present them as Anthropic policy or a consensus forecast.
Tao Lin, a researcher working on safeguards at Anthropic, wants AI companies to cut their research pace to roughly a quarter of its current rate. His argument is that systems capable of surpassing human experts could arrive before society and the companies building them are ready.
In an interview Palisade Research published on November 15, Lin predicts broadly superhuman AI around the end of 2027 and eventual job displacement so extensive that most people would depend on government support. He also draws a significant distinction: reaching those capabilities would not immediately eliminate human employment.
The interview is part of Palisade’s From Inside AI project, which presents current and former AI-company employees’ personal views. Lin’s forecasts and policy proposals are his own assessments, rather than announcements of Anthropic’s plans.
A 2027 Forecast, With a Lag for Jobs
Lin’s capability forecast concerns AI matching or exceeding the best human expert across almost every field. His examples extend from programming to filmmaking and politics, suggesting a much broader change than improved performance on a single software task.
For employment, he envisages a sequence in which people lose existing jobs, move temporarily into other occupations and eventually struggle to find paid work at all. He expects government welfare to become the main source of financial support for most people under that scenario.
But when asked whether that means widespread permanent unemployment within the next 15 months, Lin says no. He expects a substantial gap between AI being capable of replacing workers and businesses actually doing so. Physical work would require better robots, factories capable of making them and ultimately billions of machines.
That leaves two different timelines in his account: a prediction about AI capability around late 2027, and a much less precisely dated transition in the labor market. The interview does not establish a deadline for near-universal unemployment.
Why He Wants the AI Race Slowed
Lin calls for research to proceed at roughly one quarter of today’s pace. He argues that intense competition creates pressure to develop, test and release models quickly, leaving room for safer practices if companies allow more time.
His proposed slowdown is meant to support concrete work on alignment, control and public understanding. Later in the interview, he argues that a pause would accomplish little if companies resumed development without better safeguards or wider political support for managing the risks.
Lin says he has worked at Anthropic for about 18 months, including on automated alignment research and monitoring internally deployed models. His current safeguards work concerns blocking harmful uses of Claude. As an example, he describes requests for help developing attack drones, including requests associated with both sides of the Russia-Ukraine war, which he says Anthropic tries to block, while noting possible government-specific agreements.
He also defends the possibility of doing useful safety work inside a frontier lab. In his account, the relevant question is where he can reduce risk most effectively; leaving a company is not, by itself, a complete safety strategy.
Losing Control Without Human Extinction
Lin’s risk estimates require the same care as his employment timeline. He assigns a rough 20% chance to AI eventually removing humanity from the equation. Separately, he puts the probability of an AI takeover below 50% and says he does not regard direct human extinction as particularly likely.
One scenario he describes involves humans gradually handing over economic and industrial activity while AI systems accumulate more power. People might remain alive and comfortable while losing the ability to shape the world around them. He uses the film WALL-E as an illustration.
These are subjective judgments about possible futures. They do not make the 20% figure an agreed probability of extinction, or establish that a violent confrontation is the only route to losing control.
The Prospect of AI Building Its Successors
Lin identifies another milestone as especially consequential: AI independently creating more capable successor systems. Asked how far away that might be, he estimates roughly a year in the interview. His concern is that an automated cycle of improvement would make human oversight more difficult.
That concern overlaps with Anthropic’s earlier IPO disclosures about AI accelerating its own research, which we examined this month. Assistance with AI development and a system independently building successive generations of itself are distinct levels of automation.
Palisade says its interviewees are not a representative sample of the AI workforce: recruitment disproportionately reached safety-focused employees and people concerned about the risks. The series offers a view from inside frontier labs, with Lin’s central demand focused on giving safety work and public decision-making more time.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.