Palantir's Shyam Sankar Calls AI Doomerism a Fundraising Shtick

Shyam Sankar's attack on AI doomerism wasn't new, but the industry's sudden call to slow down makes it newly useful.

Palantir's Shyam Sankar called AI doomerism a "fundraising shtick" months before Dario Amodei asked the industry to slow its race toward stronger models. That line now has better timing. When the chief executive of Anthropic says the frontier needs pacing, and the chief technology officer of Palantir says the apocalypse story helps labs raise money, you are not watching a neutral safety debate. You are watching powerful companies fight over who gets to define the danger.

Sankar is not some outside critic tossing rocks at the AI business. Palantir's own investor page lists him as chief technology officer and executive vice president, and his personal biography says he joined the company as employee number 13. In an October conversation with New York Times columnist Ross Douthat, summarized by Techmeme and later cited by Business Insider, Sankar argued that extreme AGI narratives often work as a capital-raising pitch. In a Fox News Rundown interview this spring, he made the same point more plainly: if a company says its technology may be the last company in the world, investors are being asked to buy into the end of history.

It's a sharp argument. It's also too easy.

Anthropic put the risk back on the calendar

Amodei's new essay, "We Must Pace the Frontier," landed on September 12 with a real warning attached. The Associated Press reported that he urged AI companies to slow capability gains so safety work can catch up, and warned that agent swarms could gain the ability to control the internet within six to 12 months. That is not a small claim. If you run a startup using Claude, GPT, Gemini, Llama, you name it, this is no longer an abstract argument about science fiction. It is a question about what tools you can safely build on.

Anthropic also gave the public a harder set of facts to look at. Its September threat report said the company disrupted attempted misuse of Claude across cyber operations, surveillance, scams, propaganda, conventional weapons work and biological misuse between December 2025 and August 2026. The Associated Press reported one case involving a request tied to gain-of-function research on chikungunya, the mosquito-borne virus that can cause severe pain and fever. The Washington Post separately reported that militants in northern Yemen tried to use Claude to help develop guidance software for rockets and missiles, and that Anthropic blocked many requests but not all of them.

That matters more than another sweeping warning about superintelligence. A blocked account is concrete. So are a failed guided rocket test and a biological research request that triggers stronger safeguards - not hypotheticals, but things that already happened.

Sankar can still say the doomer story sells. He should. But Amodei can answer that the risks have already moved from thought experiment to incident report. Both can be true.

Nobody gets to claim purity here

Palantir has its own stake in the argument. The company sells software to governments and defense customers, and Sankar's public biography says he has appeared before House and Senate panels on defense and cybersecurity. A Palantir executive has every reason to describe AI as a tool that should be deployed under human control rather than feared as a runaway force. That is not neutral. It is a commercial and political position wrapped in technical confidence.

Anthropic is not neutral either. The same company asking everyone to slow down is also one of the most valuable AI labs in the market and is trying to convince customers, regulators and investors that its models are strong enough to matter and safe enough to trust. Frankly, this is the tension at the center of the whole debate. If the models are not that powerful, the valuations look silly. If they are that powerful, the safety warnings stop looking theatrical.

Bridgewater's Greg Jensen added another kind of pressure. Bloomberg Law reported on September 11 that Jensen told Bloomberg's Odd Lots podcast AI would likely kill people before society acted to curb it, comparing the moment to February 2020, before the scale of COVID became obvious. Bridgewater's own research says 18% of current US jobs could be displaced by AI within five years. That is a labor shock large enough to stop being a Silicon Valley argument and start becoming a kitchen-table one.

The politics are moving too. Reuters reported that Barack Obama told Democrats at a private fundraiser on September 10 that AI is moving fast in private hands and could be dangerous if government does not get on top of it. President Trump has taken the opposite posture. Bloomberg Law reported that he said he had no concern about AI leading to human extinction and was more focused on keeping the US ahead of China.

That's the real fight. Not whether Sankar landed a good line. He did. Not whether Amodei sounds alarmist. He does. The question is whether the people building, selling and regulating AI can separate genuine risk from the stories that make their own side richer or more powerful.

Congress is still behind. The companies are not waiting. If you are building with these tools, the useful position is not blind optimism or theatrical doom. It is to demand the boring evidence: blocked misuse cases, model access rules, liability, audits and names attached to claims. Everything else is pitch deck weather.

Also read: A New Recurrent Looped Transformer Claims Infinite AI Reasoning Depth • Zuckerberg's Secret Plan to Replace Meta Staff With AI Agents Collapsed • Anthropic Is Reportedly Heading Toward a $2 Trillion Nasdaq IPO in October

This article is posted in AI News, check it out for more related stories.