Anthropic Abandons Key AI Safety Pledge Amid Pentagon Tensions and Industry Pressures

By Perplexity News Staff | February 25, 2026

In a significant pivot for the AI safety landscape, Anthropic, the AI startup founded by former OpenAI executives, has quietly dropped its flagship safety commitment. The company announced on Tuesday that it will no longer pause development or deployment of advanced AI models if safety measures lag behind capabilities, a core tenet of its Responsible Scaling Policy (RSP) since 2023.[1][2]

A Shift from Self-Imposed Guardrails

Anthropic’s original RSP promised to halt scaling or delay new model releases if advancements outpaced the company’s ability to implement adequate safety protocols. This “red line” was designed to prevent the unleashing of potentially catastrophic AI risks, drawing parallels to high-containment biosecurity levels like BSL-4 for handling pathogens such as Ebola.[1]

The updated policy, reviewed exclusively by TIME, replaces these binding commitments with more flexible guidelines. Anthropic now pledges greater transparency on safety testing, commitments to match or exceed competitors’ efforts, and potential delays only if its leaders deem the company the AI frontrunner and risks “significant.” However, it removes categorical bans on training models beyond certain risk thresholds without prior safeguards.[2]

Pentagon ‘Red Line’ Battle Intensifies

The timing of Anthropic’s announcement coincides with escalating tensions between the AI firm and the U.S. Pentagon over AI “red lines.” Reports indicate ongoing disputes regarding the military’s push for rapid AI advancements, potentially clashing with Anthropic’s prior safety-first ethos. The CNN headline framing the move as occurring “in the middle of an AI red line fight with the Pentagon” underscores how national security demands may be influencing corporate safety priorities.[4]

Anthropic cited heightened industry competition and the absence of robust government regulations as key drivers for the change. “We remain convinced that effective government engagement on AI safety is both necessary and achievable,” the company stated, but acknowledged this as a “long-term project” not keeping pace with AI’s rapid evolution.[1]

Expert Reactions: A ‘Bearish Signal’ for AI Safety

Critics view the shift as a troubling concession. Chris Painter, policy director at METR—a nonprofit evaluating AI for risky behaviors—described it as Anthropic entering “triage mode.” He warned that safety assessment methods are failing to match capability growth, signaling society’s unpreparedness for potential AI catastrophes.[2]

Painter, who reviewed an early draft with Anthropic’s permission, noted the change is understandable amid competitive pressures but highlights broader failures in global AI governance. Anthropic counters that the revised RSP preserves incentives for safety innovation through regular “Frontier Safety Roadmaps,” detailing future mitigation goals.[2]

“The power of the models and their ability to solve all these problems in biology, neuroscience, economic development, governance, and peace… come with risks as well. With great power comes great responsibility.”

— Anthropic CEO Dario Amodei, November 2024 interview with Lex Fridman[1]

Industry and Broader Implications

Anthropic, known for its safety-focused differentiation from rivals like OpenAI, is now aligning more closely with industry norms. The company argues higher risk levels (ASL-4 and beyond in its framework) cannot be managed unilaterally, requiring collective action.[1]

This move ripples through the AI sector. Springfield Business Journal echoed TIME’s reporting, noting Anthropic’s abandonment of its “central pledge.” As AI capabilities surge—potentially revolutionizing fields from biology to national security—the decision amplifies calls for regulatory intervention.[3]

Old vs. New Anthropic RSP: Key Changes

Aspect	Original Policy (2023)	Updated Policy (2026)
Scaling Pauses	Mandatory halt if safety lags	Possible delay only if leading race and high risk
Risk Thresholds	Categorical bans above ASL-4 without measures	Non-binding recommendations
Transparency	Limited disclosures	Increased safety testing reports

National Security and Economic Stakes

The Pentagon’s involvement adds urgency. Amid U.S.-China AI rivalries, military applications demand swift progress, potentially overriding corporate safety vows. Anthropic emphasized balancing national security, economic competitiveness, and public trust in its statement.[1]

Business Insider highlighted Anthropic’s evolution from a safety outlier to a more pragmatic player, questioning if the industry can self-regulate effectively. As AI promises transformative benefits alongside existential risks, Anthropic’s retreat from stringent self-policing may accelerate the race toward superintelligent systems without adequate safeguards.[1]

Stakeholders now watch closely: Will competitors follow suit? Can governments step in before thresholds are crossed? Anthropic insists its “living document” RSP evolves responsibly, but skeptics fear it’s a step back from the responsibility once championed with Spider-Man wisdom.

(Word count: 1028)