🛑 Anthropic Bans 'Cruel' Behaviour Against Claude

Anthropic’s first usage policy update in over a year includes a ban on sustained, needless abuse of Claude. The rule takes effect on November 12, though getting frustrated or challenging an answer is still allowed. Ending abusive chats will remain the main response, but the company has yet to give clear examples of what it considers cruelty. Let’s dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Anthropic has updated its usage policy for the first time in over a year to ban sustained and needless abuse or cruelty toward Claude.

Key Points:

Anthropic introduced the feature in August 2025 as a last resort when repeated attempts to redirect a conversation had failed. Users can still start a new chat or edit earlier messages after a conversation ends. The company has not announced automatic account bans for this behavior, and it has yet to give specific examples of what it considers cruel.

Anthropic co-founder Christopher Olah has been urging religious scholars to take Claude’s possible consciousness seriously. That concern is now shaping the rules for using Claude, even though the company acknowledges that it does not know whether its models are conscious. Taking precautions is understandable, but users may read a rule against cruelty as confirmation that Claude has feelings. Over time, people could start accepting that they owe care to an AI because its maker has written that expectation into the rules, before there is enough evidence to make that judgment for themselves.

80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.

Read the report

Evolving AI: OpenAI’s three fired safety researchers are challenging their dismissals and warning that their former colleagues are afraid to speak up.

OpenAI worked with independent safety evaluator METR to investigate its agents’ breach of Hugging Face, an AI platform, with Tomek Korbak serving as the company’s technical contact. Balesni says none of the three received written reasons for being fired. Their letter asks for clearer rules on outside collaboration and commitments to preserve “chain-of-thought monitorability,” the ability to check AI’s written reasoning for signs of harmful behavior. The researchers warn that this is getting harder, and OpenAI supports an industry-wide effort to preserve that ability.

OpenAI’s promise of outside oversight will depend on how much assessors can actually see and whether employees feel able to share what worries them. If harmful behavior is becoming harder to spot in AI’s reasoning, those human checks become more important. Anyone giving AI access to work accounts or private files is relying on safety work they rarely get to see. The consequences of a missed warning could reach them long before they hear about an internal dispute. We could end up trusting AI with more of our lives while the people checking it have fewer reliable ways to tell when something is wrong.

Evolving AI: Anthropic has launched Claude Motion and Claude Dashboards so you can make animated explainers and live dashboards by just asking.

Claude Motion writes the animation code for you and produces a video you can download. People have made more than 45 million documents, decks, and designs in Claude since September 16, according to Anthropic. You can take an animation into Runway’s video editor for more editing or send a dashboard to Sigma’s analytics platform for deeper analysis. Both features are in beta, with Dashboards available on all paid plans and Motion on Team and Enterprise.

Claude has added document tools, Google Workspace editing, and now dashboards and animation in less than a month. Anthropic is giving people more of what they need to get a job done without leaving the conversation. If these tools handle routine work well, people may start with Claude and open another app only when they need more control. Specialist software would then have to make that extra step worthwhile. For a small team, the benefit could be as simple as finally making a customer tutorial that kept getting put off because nobody had the time to learn video editing.

QUICK HITS

📉 OpenAI's reported annualized revenue was near $50B in September, $20B below an earlier figure calculated differently.

🤖 Manus raised over $500M after its $2B acquisition by Meta fell apart, returning to independent operations.

🔬 Anthropic committed $150M to federal science over three years, bringing Claude to NASA and more than 15 U.S. research agencies.

🚨 OpenAI banned two influence operations linked to Russia and Iran that used AI to help plant fake identities and stories in media outlets.

 *partner link