AI · Policy · October 14, 2026

An Anthropic Executive Says Open Models Carry “More Risk.” Here Is the Company’s Actual Position, and Who Disagrees.

On Oct. 14, 2026, Guillaume Princen, Anthropic’s head of international, told Bloomberg TV that open-weight AI models carry more risk than closed ones. “We need to be really careful with open-source on safety,” he said at the International AI Summit in Dublin, per The Next Web and IBTimes Singapore. “By definition, there’s more risk with them.”

The remark landed in a live fight. Nvidia, Hugging Face, Palantir and others argue open models are a safety asset and that Anthropic wants them restricted. Anthropic’s own written position is narrower than the headline.

Oct. 14, 2026 — Princen's Bloomberg TV remarks at the International AI Summit in Dublin · Source: The Next Web; IBTimes Singapore
Never a ban — Anthropic says it "has never advocated for a ban on open-weights models" · Source: Anthropic position post, July 27, 2026
All models — Anthropic's proposal: mandatory safety testing for every sufficiently capable model, open and closed · Source: Anthropic position post
Not signed — Anthropic did not sign the July 24 open-models letter that Nvidia signed · Source: Fortune

Disclosure

Anthropic is a party to this story: it is the speaker, it sells closed models, and its Claude models are used in this site’s production workflow. We have sourced every claim below to documents or outside outlets so readers can weigh it themselves.

§ 01 / What Princen Said

Bloomberg’s article is paywalled and we could not open it, so the quotes here come from The Next Web and IBTimes, which both cite the TV interview. The Next Web renders the fuller line as “I can’t talk for open models. But by definition, there’s more risk with them.” Bloomberg’s headline says “higher risk”; Princen’s reported wording is “more risk.”

Neither outlet quotes Princen calling for a ban or for rules on open models. The Next Web says he went further than Amodei’s July position; that is the outlet’s reading, not an Anthropic statement.

Keep Reading

AI · Security

A Saudi National at OpenAI Is Accused of Hiding a Recording Device in an Executive Conference Room

Read →

§ 02 / The Written Position

CEO Dario Amodei laid it out on July 27. Open-weight models “do potentially present a higher risk than closed models” because it is difficult to apply guardrails or monitor their use, the post says. But: “Anthropic has never advocated for a ban on open-weights models.” Its proposal is that “all sufficiently capable models, open and closed, should go through mandatory safety testing.”

Amodei also conceded the competitive point critics raise: a ban “would protect US AI companies from competition, but that has never been my goal,” per the Daily Caller. Anthropic’s Frontier Red Team published related research on Oct. 12 under the title “AI models may be capable of self-replicating without access to a datacenter.” IBTimes describes tests in which open-weight models coordinated across devices in a firewalled environment, so they do not show a worm loose on the internet.

Cartoon satire: a workbench with a glowing cordless drill in an open toolbox beside a stack of blank booklets, a blank sticky note and safety goggles, with a pegboard of tool outlines behind — The toolbox question: a powerful tool left on the bench with a blank instruction sheet. Who should be allowed to take it is the whole debate.

§ 03 / The Case Against Restrictions

In his first X post, Nvidia CEO Jensen Huang shared a letter Nvidia signed on why open models matter; Fortune quotes him saying “the world needs both frontier closed models and frontier open models.” Palantir CEO Alex Karp called a plan that restricts AI products for U.S. commercial use but not abroad “complete insanity,” per Fox Business. David Sacks, described by PPC Land as an outside White House AI adviser, said closed labs “want the government to eliminate their open-source competition,” as quoted by PPC Land.

Jensen Huang

@JensenHuang · July 24, 2026

For my first post, I'm sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

Hugging Face CEO Clem Delangue, interviewed by Bloomberg Tech, argued that a “too dangerous” label is good marketing. The Next Web reports Hugging Face used Z.ai’s open GLM model for forensics after a breach, when commercial tools declined to analyze the attacker’s code. That is one incident, not a trend.

Bloomberg Tech — Hugging Face CEO Clem Delangue on Anthropic's 'dangerous' label (not the Princen interview)

Keep Reading

Tech Intelligence · AI Infrastructure

One AI Data Center, About $30 Billion. Most of It Is the Chips.

Read →

§ 04 / Interest, Evidence, and Policy

The interest question is fair to ask. Anthropic earns revenue selling access to proprietary models, and Constellation Research editor Larry Dignan notes open weights could cut into that token revenue. Against that, Anthropic’s testing proposal applies to its own closed models too, and it did not call for a ban.

The evidence is thin on both sides. The risk case leans on tests by Anthropic and safety groups; the benefit case leans on incidents like the Hugging Face forensics. On policy, the White House’s July 2025 AI Action Plan encourages open-source and open-weight models, while the administration has reportedly weighed limits on Chinese open models, per the Daily Caller citing Axios.

What We Do Not Know

The full text of Princen’s Bloomberg interview, and his exact title; outlets agree on “head of international.”

Whether Anthropic will seek binding rules on open-weight releases.

Whether any open model has yet caused a serious real-world harm that testing would have caught.

“Anthropic has never advocated for a ban on open-weights models.”
Anthropic, 'Our position on open-weights models,' July 27, 2026

The bottom line: an Anthropic executive said open models carry more risk; Anthropic’s written position says the same while ruling out a ban and asking for testing of everyone. Critics call that self-serving, and Anthropic says it is not. Readers have the quotes, the documents, and the disclosure to decide.