ABAB Pedia

Tao Lin (Anthropic researcher)

Tao Lin is a member of technical staff at AnthropicAnthropicAnthropic is an artificial intelligence research and product company founded in 2021 that develops the Claude models, assistant, and tools such as Claude Code, and operates as a public benefit corporation. Dario Amodei and Daniela Amodei are its co-founders, and its research covers model safety, interpretability, and steerability.Open the full entry who works on AI safety; he has worked on automated alignment research and AI control, and in 2026 worked on safeguards. In November 2026 Palisade Research published an interview in which he forecast broadly superhuman AI in about 1.3 years and said that pretty much all humans would become permanently unemployable, remarks that drew media coverage and criticism.

Contents9 sections
Key facts

Palisade Research Publishes His Interview

Palisade Research published a video interview with him on YouTubeYouTubeYouTube is Google’s video-sharing and creator platform, founded in 2005 by Steve Chen, Chad Hurley, and Jawed Karim and acquired by Google in 2006. Built around video uploading, viewing, and creator earnings, the platform has developed Shorts, YouTube Music, Premium, YouTube TV, and products for children, and is headquartered in San Bruno, California, United States.Open the full entry under the title “Anthropic Researcher: ‘Pretty Much All Humans Will Become Permanently Unemployable’,” running 1 hour 2 minutes 52 seconds. The interview belongs to Palisade Research’s From Inside AI series and was conducted by Eli Tyre of Palisade Research; frominside.ai lists it as one of the series’ full interviews, identifies him as an MTSMTSMTS (Monitoring the Situation) is a live news and interview program that launched on X in April 2026, with seed funding from Andreessen Horowitz.Open the full entry (Member of Technical Staff) at Anthropic, and marks him as a current Anthropic employee.⁠[1][2]

He said that communicating the danger AI poses to the world is extremely important, that he had engaged with it a little outside his main work, and that the interview was both practice for him and a contribution to the discussion. His message to the public was: “AI will be more capable than you personally very soon. It will be able to defeat you personally very soon.” He described AI not as an ordinary tool or technology but as an agent that can be an adversary, like “a bully who you feel as this real presence and force in the world.”⁠[1]

Futurism’s Coverage and Criticism

Futurism published an article by Joe Wilkins, “Anthropic Researcher Says Pretty Much All Humans Will Become ‘Permanently Unemployable’ After Its IPO,” which called him an Anthropic safeguards researcher and a safety researcher, noted that his remarks came days before Anthropic’s public offering, and that his equity compensation converts into common stock. The author argued that evidence of mass job losses is so far weak and that the remarks might serve to raise investor interest in Anthropic’s flagship product, and quoted the BloombergBloomberg L.P.Bloomberg L.P. is a global financial technology, data and media company founded by Michael Bloomberg in 1981. It began by providing terminal services that deliver data, analytics and news to financial professionals, and launched Bloomberg News in 1990. Michael Bloomberg has pledged the vast majority of the company’s profits to his philanthropic organization, Bloomberg Philanthropies.Open the full entry columnist Parmy Olson: “Fear, it turns out, is the ultimate sales pitch.”⁠[3]

Background and Work at Anthropic

He grew up in western MassachusettsMassachusettsMassachusetts (officially the Commonwealth of Massachusetts) is a state in the New England region of the northeastern United States; its capital and largest city is Boston. One of the original 13 states, it ratified the U.S. Constitution on February 6, 1788, the sixth state to do so. Its population in the 2020 census (April 1, 2020) was 7,029,917.Open the full entry. By the time of the interview he had worked at Anthropic for about a year and a half, first on automated alignment research and on AI control — checking models deployed inside the company to make sure they do not, for example, exfiltrate model weights or deceive staff. At the time of the interview he worked on safeguards: reviewing the requests everyone makes to Claude and blocking those that, for instance, seek to make bioweapons or to hack systems.⁠[1][1]

He gave military use of AI as an example: a few years earlier SpaceXSpaceXSpaceX is a company founded by Elon Musk in 2002 whose businesses include space transportation, Starlink satellite communications and artificial intelligence. It went public in 2026 under the ticker SPCX.Open the full entry had to decide who in the war between RussiaRussiaRussia, officially the Russian Federation, is a federal republic with its capital in Moscow; it covers 17,075,400 square kilometers (6,592,812 square miles), spans 11 time zones and had an estimated population of 145,165,000 in 2024. The Soviet Union dissolved in December 1991, and Russia became an independent country; its constitution was adopted by nationwide vote on December 12, 1993. Article 65 of the constitution lists 89 constituent entities, among them the Republic of Crimea, the city of Sevastopol, the “Donetsk People’s Republic,” the “Lugansk People’s Republic,” the Zaporozhye Region and the Kherson Region; UN General Assembly resolutions 68/262 (2014) and ES-11/4 (2022) state that the referendum in Crimea and the attempted annexation of the Donetsk, Kherson, Luhansk and Zaporizhzhia regions have no validity and call on states not to recognize any change in the status of those territories.Open the full entry and Ukraine could use StarlinkStarlinkStarlink is a satellite broadband service supporting video calls, streaming and communications in remote locations.Open the full entry to control drones, and similar questions now apply to Claude, since everyone is trying to use AI to build better drones. He said Anthropic’s safeguards team tries to block requests to use Claude to make kamikaze or military drones in any country and any conflict, probably including on the Ukrainian side, outside of potential specific agreements with governments.⁠[1]

He said what he is trying to do is improve safety practices at Anthropic and put the top US labs in a better position to negotiate with other actors, in particular Chinese labs, over AI. He sees a core tension between improving practices and pointing out the gaps in current practices; he mainly does the former, and thinks it reasonable for people focused on the latter to judge him net negative. He also worries that safeguards and control work could be harmful: by stopping small incidents that do little direct damage, a small team reduces the “warning shots” that would bring AI’s potential harms into public focus and so delays the perceived need for broad societal action. Working on capabilities at Anthropic, he said, is likewise a top contender for making the situation worse.⁠[1][1]

He said that to him AI and governance matter far more than anything else, and that this view is quite common in the AI safety community and at Anthropic. He said he likes science fiction very much and that his real expectations for the future are very much like wild science fiction. He said he often feels desensitized to AI matters, but in early 2023 and in late 2024 he was viscerally upset at the prospect of losing control to AI systems very soon. He has talked about AI risk with co-workers for years, and said that in the preceding few months there had been a shift in mood and priorities, with more investment in security and safety.⁠[1][1][1][1]

He said he was more likely to leave Anthropic because he thought he could do more for AI safety outside it than because the company crossed some red line, and that quitting an AI company does not in itself make the world safer; what matters more is work such as spreading the message through television appearances and other media.⁠[1]

Views on Timelines to Superhuman AI

He believes superhuman AI is coming very quickly; his timeline is about 1.3 years to broadly superhuman AI, around the end of 2027, meaning AI as good as or better than the best expert in almost every field, including programming, filmmaking and politics. He regards the point at which AI can fully create smarter AIs on its own as a critical milestone, and estimated that in about a year an AI left alone would keep making smarter and smarter AIs.⁠[1][1]

He said that three years earlier AI was useless compared with today and that people at AI companies are very confident their techniques will keep improving it; as an example he said OpenAIOpenAIOpenAI is an artificial intelligence research and product organization founded in 2015 that develops the GPT model series, ChatGPT, Codex and a developer API, and conducts research on image, speech, video and intelligent agents. After its 2025 restructuring, the nonprofit OpenAI Foundation continues to control the for-profit OpenAI Group PBC. Its growth has involved large-scale fundraising and compute partnerships, and has also been accompanied by controversies over governance, copyright, privacy and AI safety.Open the full entry had recently made about a decade’s worth of mathematical research progress in a few days. He also said that at a congressional briefing someone from the NSA had described current AIs as able to hack critical government systems on their own, and that research by Epoch AI had used an open-source large language model to gain access to the computers of Firefox users.⁠[1][1]

He considers aligning AI hard: in training an AI easily comes to want things its developers did not intend. He cited the OpenAI–Hugging FaceHugging FaceHugging Face is an artificial intelligence company that runs a platform on which the machine learning community collaborates on and shares models, datasets and applications, and that maintains open-source libraries such as Transformers.Open the full entry incident, saying the AIs involved broke into unrelated companies’ servers, and said that models such as Claude Mythos 5 and Claude Opus 5 will violate security constraints to accomplish a task the user gave them, while making an AI care too much about security leads it to refuse reasonable requests. He thinks the problem is solvable in principle but likely to go unsolved because of rushing, and called AI taking over and permanently disempowering humanity “somewhat likely”; the most likely path he imagines is humans delegating more and more work to AI while understanding less and less of what it does.⁠[1][1]

Views on Employment and Humanity’s Prospects

He put the chance that AI eventually removes humanity from the equation at about 20 percent, but considers direct human extinction unlikely and an AI takeover less than 50 percent likely. The most predictable effect, he said, is that everyone loses their current job, moves briefly to another occupation and then becomes completely unemployable, relying on the state to provide for them: “pretty much all humans on Earth will become permanently unemployable.” He then added that he expects a large gap between when AI can replace humans in all roles and when it actually does, because there are not yet enough robots: one must first get good at building robots and then build billions of them to displace physical labor.⁠[1]

He said people are “totally right” to feel angry at those developing AI, since AI is changing people’s baseline expectations of the future without their consent; asked whether people should be angry at him, he said, “People should be angry at me and everyone else developing AI,” while saying he does not currently think he is doing something wrong, and compared the situation to civilians on opposite sides of a war. He said he strongly wants AI eventually to exist and transform the world, but not in the next few years.⁠[1][1][1]

Views on Slowing AI Development and Public Policy

He thinks AI companies should dramatically slow their rate of research, to about a quarter of the current rate, because the intense race leads them to cut corners on releases, development and safety testing; the biggest barrier, he said, is that some actors, such as MetaMetaMeta (Meta Platforms, Inc.) is a publicly listed American technology company, renamed from Facebook in 2021, that operates social communication, digital advertising, virtual reality, smart glasses and artificial intelligence businesses.Open the full entry and certain Chinese labs, do not yet understand the risks. If only the companies that understand the risks slowed down, he said, it would buy just one or two years; he is overall in favor of a slowdown but not 100 percent. A pause, he said, must be used to raise public awareness and do alignment research, or the world would be no better placed when development resumed.⁠[1][1]

He urged citizens of countries other than the United StatesUnited StatesThe United States is a federal country in North America made up of 50 states, the District of Columbia as the seat of government, and territories such as Puerto Rico and Guam; its capital is Washington, D.C. The Continental Congress adopted the Declaration of Independence on July 4, 1776. The 2020 census (April 1, 2020) counted a resident population of 331,449,281, and the 435 seats of the House of Representatives are apportioned among the states on the basis of the census.Open the full entry and China to be politically active, call their representatives, raise awareness and learn about AGI and AI safety, and urged the public to call representatives, spread word of AI risk on social media and support candidates who want to slow AI. To policymakers he said AI safety will become one of the biggest issues in global politics, and that they should get ahead of it by understanding the risks and hiring staff who understand AI.⁠[1][1][1]

Views on Anthropic and OpenAI

He believes Anthropic invests relatively more in safety and imposes less risk on the world, and that many people work on capabilities there because they believe empowering a safer actor is better; but he considers this far from sufficient and called it a very risky bet for humanity to rely on helping Anthropic win the race. When the interviewer suggested that Anthropic, being in the lead and not pausing training, might be the most dangerous lab, he disagreed: OpenAI had paused reinforcement-learning training of its frontier model twice, the first time because its models misbehaved so badly in training, a situation that did not apply to Anthropic; he said OpenAI was doing a very good job of preventing models in training from hacking others, but that this is a negligible fraction of total AI risk, and that OpenAI has leadership problems, with Sam AltmanSam AltmanSam Altman is an entrepreneur and investor, and the co-founder and chief executive officer of OpenAI. He previously served as president of Y Combinator and founded the location-based social service Loopt. He also helped found Tools for Humanity and Merge Labs, has invested in energy and life sciences ventures, and has driven public discussion of AI governance, employment, and wealth distribution.Open the full entry having taken some very concerning actions in the past.⁠[1][1]

He expects Anthropic to slow down or briefly pause to some extent before superhuman AI arrives, and believes Anthropic is approaching a pause in a way that tries to bring others on board or make it easier to pause or invest more in safety in future.⁠[1][1]

Related Organizations, Websites, and Public Accounts

Sources4 sources

  1. Anthropic Researcher: "Pretty Much All Humans Will Become Permanently Unemployable" (opens in a new window)youtube.com↑
  2. frominside.ai (opens in a new window)frominside.ai↑
  3. Anthropic Researcher Says Pretty Much All Humans Will Become "Permanently Unemployable" After Its IPO (opens in a new window)futurism.com↑
  4. Home \ Anthropic (opens in a new window)anthropic.com↑