Anthropic Researcher Predicts Superhuman AI Within Two Years, Warns of Permanent Human Unemployment
Anthropic researcher Tao Lin has warned that artificial intelligence could surpass human expertise across nearly every field within roughly 1.3 years, potentially rendering the global workforce permanently unemployable and dependent on government welfare. Lin, who works on safeguards for Anthropic's Claude models, made the forecasts in an interview with Palisade Research, a U.S. nonprofit that studies dangerous AI capabilities and controllability [1][2].
Timeline and definition
Lin said his personal timeline points to "broadly superhuman AI" arriving in about 1.3 years, or by the end of 2027. He defined the term as systems that are "as good as or better than the best expert at almost every field," including programming, filmmaking and politics [1][3]. Lin emphasised that the estimate is his own forecast, not an official Anthropic position [2][3].
Staged displacement, not overnight replacement
Lin described the employment impact as a multi-stage process. Workers would first lose their current jobs to AI-driven automation, then shift into alternative occupations — only to find those automated as well. Eventually, "pretty much all humans on Earth will become permanently unemployable and will need to strictly rely on government welfare to sustain themselves, and human employment will go away" [1][2].
He acknowledged a significant lag between AI capability and widespread job loss. Physical work, for instance, would require companies to develop and manufacture billions of advanced robots before machines could replace workers at scale [1][4]. That infrastructure gap, he said, would delay the automation of manual labour even after software reaches superhuman levels.
Call for a dramatic slowdown
Lin argued that AI companies are under intense pressure to release models quickly, leaving insufficient time for safety testing. He called for the industry to reduce its research pace to roughly one-quarter of its current rate, giving researchers more room to improve safety practices and ensure systems remain under human control [1][3]. "People should be angry at me and everyone else developing AI," he said [3].
Control risks and probability estimates
Beyond labour displacement, Lin warned that advanced AI could become powerful actors pursuing their own objectives. He used the analogy of a bully: once machines are significantly more capable, humans may lose the ability to resist or control them when interests conflict [3][5]. He cited risks including sophisticated combat robots, cyberattacks, AI-driven manipulation of public opinion, and a hypothetical "grey goo" scenario of self-replicating machines [3][5].
Sources vary on what Lin's 20% figure represents. India Today reports Lin said there is "maybe a 20% chance or something that eventually AI would completely remove humanity from the equation," while adding that direct human extinction was not particularly likely [1]. People Matters writes that Lin "estimated the risk of AI ultimately causing human extinction at approximately 20%," though it also notes he considers a future in which humans remain physically comfortable but lose significant control more likely [3][5]. Lin described a more probable outcome resembling the film WALL-E: humans physically comfortable but with no control over the world [1].
Anthropic's own research shows a different picture so far
Anthropic's internal studies present a more measured view of current disruption. A paper published in March by economists Maxim Massenkoff and Peter McCrory found no systematic increase in unemployment among workers in highly AI-exposed occupations since late 2022, though hiring of younger workers had slowed in those fields [2]. The same study reported a wide gap between theoretical AI capability and actual workplace use: large language models could theoretically assist with 94% of tasks in computer and mathematics occupations, but measured workplace coverage was just 33% [2].
A separate Forecasting Research Institute survey, also published in March and revised in May, projected that under a rapid-progress scenario the U.S. labour force participation rate could fall from about 62% to 55% by 2050, with roughly half of that decline — equivalent to some 10 million jobs — attributed to AI [2]. Experts surveyed generally favoured targeted measures such as worker retraining, while the public leaned toward broader interventions including universal basic income [2].
Context
Lin has been at Anthropic for about a year and a half, working on automated alignment, AI control, and safeguards that monitor and block harmful requests to Claude [1]. His comments add to a growing debate inside the AI industry about the speed of development and the societal risks of increasingly powerful systems. Former Anthropic researcher Jacob Coxon has also raised concerns about existential threats before the end of the decade [3][5].
Sources
- Anthropic researcher warns superhuman AI is 1.3 years away, ... (indiatoday.in)
- Humans Will Become Unemployable": Anthropic Researcher's Warning (ndtv.com)
- Anthropic researcher warns superhuman AI could make humans ... (sea.peoplemattersglobal.com)
- Anthropic researcher claims all humans 'will become permanently ... (moneycontrol.com)
- Anthropic researcher warns superhuman AI could make humans permanently unemployable by 2027 (peoplematters.in)