Anthropic has released an updated version of its Claude 3.5 Sonnet AI model, introducing a significant new capability called “computer use.” This feature allows the artificial intelligence to interact with computer interfaces much like a human, including moving cursors, clicking buttons, and typing text. While the company highlights the potential for automating complex, multi-step workflows and increasing productivity, the technology has prompted discussions regarding security. Due to concerns about potential misuse in cyberattacks or disinformation campaigns, Anthropic is implementing specific safety measures and a staggered rollout for the feature.
- The new “computer use” capability enables Claude 3.5 Sonnet to navigate desktop environments and web browsers to complete tasks autonomously.
- Developers can use the model to automate repetitive administrative duties, such as filling out forms, conducting research, and managing software applications.
- Safety experts have raised concerns that the ability for AI to control computers could be exploited for malicious activities, including advanced phishing or system intrusions.
- Anthropic has integrated safeguards to prevent the model from engaging in high-risk actions, such as social media interaction or voting, during its initial release phase.
- The release places Anthropic in direct competition with other AI developers, such as OpenAI and Google, who are also pursuing autonomous agent technology.
France 24 is an international television network and news website owned by the French state.
Official website: https://www.france24.com/en/
Original video here.
This summary has been generated by AI.
![Pinterest](https://pinterest.com/pin/create/button/?url=https://allnewschannels.com/news/france-24/2026/04/14/anthropic-limits-release-of-new-ai-model-citing-safety-risks-and-advanced-capabilities/&media=https://allnewschannels.com/wp-content/uploads/2026/04/Anthropic_Limits_Release_of_New_AI_Model_Citing_Safety_Risks_and_Advanced_Capabilities.png&description=Anthropic launches its new "computer use" AI capability for Claude 3.5 Sonnet, implementing a staggered, safety-focused rollout to address potential security risks.){kind=link}
Launch it at china😂
Sounds like it can convert machine code to Python at speed. I'm not sure that qualifies as "too powerful to release". More like a bit of marketing.
Ummmm you all know that many countries are building these ai tools. They are essentially electronic WMD’s
When AI threatens the money, suddenly 'too dangerous'!
Meh. Sounds like one of those stories AI companies leak to boost their own stock.
They aren't releasing because they lack the infrastructure to handle high usage volumes.
Never mind Trump. I’m pretty sure it’s AI that will blast the world back to the Stone Age.
It's just marketing nonsense. We already have tools that can find flaws in software. We don't keep them from the world. If they can find flaws, that's good for the world.
an amazing marketing moves.
they were so concerned with if they could do it they never asked if they should do it. (I came up with that quote all on my own from watching Jurassic Park)
Neuromancer