Anthropic employees accuse Trump administration of targeting them
Anthropic’s claims that its Mythos/Fable AI models are unusually powerful and potentially dangerous have collided with the Trump administration’s decision to impose sudden export-style controls that block access for non‑US users and many of the company’s own staff. Commenters debate whether this is justified national security oversight or politically motivated retaliation for Anthropic’s refusal to support US weapons systems and its public calls for stricter AI regulation. The episode is seen as a warning about arbitrary, uneven regulation, prompting worries about reliance on US-based proprietary AI and renewed interest in open‑weight and non‑US models.
Nature of the government action
- Many commenters see the export controls on Mythos/Fable as arbitrary, personalized punishment rather than coherent regulation.
- Others argue that any government order constraining a company is, by definition, “regulation,” even if badly designed or inconsistently applied.
- Several stress that regulation normally implies clear, general rules; a one-off, opaque order is framed as political interference, not policy.
Anthropic’s prior calls for regulation and “danger” framing
- Repeated references to Anthropic’s own messaging: Mythos as uniquely powerful and dangerous, initial refusal to broadly release it, calls for strong government authority to block “unsafe technology.”
- Some say Anthropic “asked for this” by fear-based marketing and quasi-regulatory-capture proposals (e.g., heavy rules for large labs).
- Others counter that Anthropic wanted a transparent, industry-wide framework, not to be singled out while competitors like GPT‑5.5 continue unaffected.
Fable guardrails and the Amazon/jailbreak episode
- Disagreement over what Amazon actually did:
- One side claims Amazon “removed guardrails,” effectively accessing Mythos and proving Fable’s safeguards are weak.
- The other side says Amazon simply asked Fable to “fix bugs” in insecure code, which is routine and not equivalent to disabling security guardrails.
- It remains unclear from the thread exactly how broad or serious the jailbreak/bypass was, and participants note that reports and parties involved are not fully trustworthy.
Targeting, retaliation, and rule-of-law concerns
- Strong theme that Anthropic is being targeted for:
- Refusing US defense/intelligence requests for broader access or weapons integration.
- Prior political/public criticism of the administration.
- Some see this as emblematic of a “loyalty over law” executive style, with selective enforcement to reward allies and punish dissenters.
Implications for AI ecosystem and users
- Many say this demonstrates that relying on US closed models is an existential business risk; capabilities can be restricted overnight.
- Growing enthusiasm for open-weight models and non‑US hosting as a hedge against US export controls.
- Some argue China’s open-weight push and other countries’ moves will accelerate as trust in US-based AI providers erodes.
Attitudes toward Anthropic
- Mixed reactions: sympathy over arbitrary state power vs. schadenfreude over perceived regulatory-capture attempts backfiring.
- Recurrent view: “Yes, they’re being targeted—and they also helped create the narrative that enabled it.”