Anthropic weakens its safety pledge in the wake of the Pentagon's pressure campaign

Published: (February 25, 2026 at 01:34 PM EST)
5 min read
Source: Engadget

Source: Engadget

Two Stories About Anthropic That Together Paint a Chilling Picture

1. Pentagon Pressure on Anthropic

“If one AI developer paused development to implement safety measures while others moved forward training and deploying AI systems without strong mitigations, that could result in a world that is less safe.” – Excerpt from Anthropic’s revised Responsible Scaling Policy (RSP)


2. Anthropic Lowers Its Safety Guardrails

On the same day the Pentagon story broke, Anthropic announced a revision to its Responsible Scaling Policy (RSP), effectively lowering safety guardrails that had previously halted model training unless strict safeguards were guaranteed.

  • Original pledge: Stop training new AI models unless specific safety guidelines could be guaranteed in advance.
  • New approach: Replace hard “tripwire” red lines with “Risk Reports” and “Frontier Safety Roadmaps,” offering transparency rather than firm limits.

Source: Anthropic – “Responsible Scaling Policy v3”


Anthropic’s Own Words

In an exclusive interview with Time, Anthropic’s Chief Science Officer Jared Kaplan explained the shift:

“We felt that it wouldn’t actually help anyone for us to stop training AI models. We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments… if competitors are blazing ahead.”

Source: Time interview (via Yahoo Shopping link)


Anthropic CEO Dario Amodei (Photo by David Dee Delgado/Getty Images for The New York Times)

Anthropic CEO Dario Amodei


Context: Valuation, Competition, and Public Perception

  • Valuation: Anthropic raised $30 billion in a Series G round, pushing its post‑money valuation to $380 billion.
  • Competitors: OpenAI is currently valued at over $850 billion.
  • Product Praise: The latest Claude models have been lauded for coding assistance and overall safety.
    • Claude Sonnet‑45: “Anthropic’s safest AI model yet.” – Engadget
    • Claude Co‑Worker: Coding AI for regular people. – Engadget

These factors help explain why Anthropic might feel pressure to stay competitive while also navigating government demands.


New Safety Framework

Old MechanismNew Mechanism
Hard “tripwire” red lines that halt developmentRisk Reports – public disclosures of safety assessments
Fixed safety guardrailsFrontier Safety Roadmaps – evolving safety plans aligned with industry progress

The revised RSP cites a “collective action problem”: if only a few developers pause for safety while others race ahead, the overall ecosystem becomes less safe.

Source: Anthropic – Revised RSP excerpt


Defense Secretary Pete Hegseth (Photo by Aaron Ontiveroz/The Denver Post)

Defense Secretary Pete Hegseth


The Elephant in the Room: Pentagon Pressure

Neither Anthropic’s announcement nor the Time interview mentions the Pentagon’s pressure campaign.

  • Axios report (Feb 24, 2026): Hegseth told Anthropic CEO Dario Amodei the company has until Friday to give the military unfettered access to Claude or face penalties.

  • Anthropic’s response: The company is willing to adopt its usage policies for the Pentagon but refuses to allow:

    1. Mass surveillance of Americans.
    2. Weapon systems that fire without human involvement.
  • Potential penalties:

    • Invocation of the Defense Production Act, allowing the president to direct private firms to prioritize defense contracts.
    • Possible termination of the Pentagon contract and designation of Anthropic as a non‑compliant vendor.

If Anthropic resists, experts suggest legal action may be its best recourse, though the effectiveness of the Pentagon’s threats against a profit‑driven startup remains uncertain.


Bottom Line

  • Anthropic’s policy shift reduces hard safety limits in favor of more flexible, disclosure‑based mechanisms.
  • Simultaneously, the Pentagon is pressuring the company for unrestricted military access to Claude.
  • The convergence of commercial competition, government demand, and softening safety commitments creates a potentially risky landscape for AI development and oversight.

Supply Chain Risk

A supply chain risk would force other companies working with the Pentagon to certify that Claude isn’t included in their workflows.

Claude is the only AI model currently used for the military’s most sensitive work.

“The only reason we’re still talking to these people is we need them and we need them now,” a defense official told Axios. “The problem for these guys is they are that good.”

Claude was reportedly used in the Maduro raid in Venezuela, a topic Amodei is said to have raised with its partner Palantir.


Anthropic’s Revised Safety Plan (RSP)

Time’s story about the new RSP included reactions from a nonprofit director focused on AI risks. Chris Painter, director of METR, described the changes as both understandable and perhaps an ill omen.

“I like the emphasis on transparent risk reporting and publicly verifiable safety roadmaps,” he said.

However, he also raised concerns that the more flexible RSP could lead to a “frog‑boiling” effect. In other words, when safety becomes a gray area, a seemingly never‑ending series of rationalizations could take the company down the very dark path it once condemned.

Painter said the new RSP shows that Anthropic “believes it needs to shift into triage mode with its safety plans, because methods to assess and mitigate risk are not keeping up with the pace of capabilities. This is more evidence that society is not prepared for the potential catastrophic risks posed by AI.”


If you buy something through a link in this article, we may earn commission.

0 views
Back to Blog

Related posts

Read more »