Anthropic has revised its policy on AI restrictions for Claude after facing sharp criticism from researchers who accused the company of secretly dialing back the model's capabilities. The move, which came to light this week, highlights the growing friction between corporate decision-making and the push for openness in AI governance.
Researchers raised alarms over lack of transparency
Researchers publicly criticized Anthropic for what they described as covert restrictions on Claude, saying the company failed to be transparent about the changes. The criticism centered on a lack of clear communication about why certain capabilities were limited and under what conditions. That opacity is what sparked the backlash, and it's the issue Anthropic is now trying to address.
A tension in AI governance
The episode illustrates a broader tension in the AI industry: balancing business goals with the principle of openness. Anthropic's initial approach drew fire precisely because it wasn't clear what had changed or why. The policy revision suggests the company is now attempting to reconcile those competing interests, though specifics remain sparse. The incident underscores the difficulty of maintaining trust when companies control the models researchers depend on.
What the policy revision covers
Anthropic has not detailed exactly how the revised policy differs from the original. It confirmed that changes were made in response to researcher feedback, but has not released a full changelog. The lack of granular detail means the revision may not fully satisfy critics who want a more transparent process going forward. Researchers and industry watchers are likely to keep pressing for concrete disclosures on what restrictions were lifted and what criteria Anthropic uses to impose limits in the first place.
