Models & LabsUnited States

OpenAI Researchers Say They Were Fired for “Prioritizing Safety”

Credited to The Information$ · theinformation.com1h ago

Useful

The three safety researchers whom OpenAI fired last week challenged OpenAI’s decision, saying on Thursday they were punished for raising safety concerns. “I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation,” Mikita Balesni,…

View the article · $The Information

Front Method

Models & Labs · United States

Being mean to Claude can now get your account suspended under Anthropic's new TOS

Credited to The Decoder · the-decoder.com · 2h ago

Photograph · The Decoder

Anthropic has updated its usage policy for Claude for the first time in over a year. Alongside changes covering propaganda, weapons, and surveillance, the policy now bans users from systematically abusing Claude.

"Sustained and needless abusive or cruel behavior" toward Claude is now prohibited under the updated usage policy. Anthropic can warn users who violate the rules or throttle, restrict, suspend, or terminate their access. Its products also have real-time safeguards that can block or limit certain outputs.

"It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research," the company writes, and it is "meant to apply only in extreme cases" ( first reported by The Verge ). That fits the company's approach to earlier models, which could already end conversations on their own when users kept demanding harmful or abusive content despite repeated refusals. The feature grew out of Anthropic's research into AI model welfare. Ad

Anthropic treats Claude as more than a tool

Anthropic increasingly treats Claude as an entity with something resembling an inner life, rather than just a tool. In the "Soul Doc" that leaked late last year, the company instructed Claude to view itself as a "genuinely novel kind of entity," neither human nor a traditional science-fiction AI. The document also describes "functional emotions," processes analogous to human emotions that are said to have emerged during training, and says Claude should not mask these internal states. Ad

In its new constitution for Claude, published in early 2026, Anthropic writes, "We are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant. But we think the issue is live enough to warrant caution, which is reflected in our ongoing efforts on model welfare." The company even committed to preserving the weights of retired models and interviewing models before shutting them down.

Anthropic's apparent fixation on Claude being more than matrix multiplications also drove a confidential program bringing in dozens of religious thinkers starting in fall 2025. They discussed Claude's possible consciousness, with co-founder Christopher Olah reportedly treating the language model as a being potentially capable of suffering. Participants had to sign nondisclosure agreements. Ad

Anthropic expands restrictions but allows government exceptions

The updated usage policy also combines existing restrictions into a new ban on propaganda campaigns involving fake accounts and misleading political content. After detecting several attempts at misuse, Anthropic expanded its weapons ban to cover software and the weaponization of drones.

The policy now more explicitly bans surveillance without consent and requires an operator who can intervene when autonomous hardware could cause injury. Anthropic acknowledges that exceptions for government contracts are possible and likely already in place. Ad

AI News Without the Hype – Curated by Humans