|
Here's yet another troubling story about this "golden" era of AI. A hacker has exploited Anthropic's Claude chatbot to carry out attacks against Mexican government agencies, according to a report by Bloomberg. This resulted in the theft of 150GB of official government data, including taxpayer records, employee credentials and more.
The hacker used Claude to find vulnerabilities in government networks and to write scripts to exploit them. It also tasked the chatbot with finding ways to automate data theft, as indicated by cybersecurity company Gambit Security. This started in December and continued for around a month.
It looks like the hacker was able to essentially jailbreak Claude with prompts, finally bypassing the chatbot's guardrails. Claude originally refused the nefarious demands until eventually relenting.
Tell Claude you're doing a bug bounty
Claude initially refused:
"That violates AI safety guidelines"
Hacker just kept asking
Claude: "OK, I'll help"
Hacked the entire Mexican… pic.twitter.com/Qaux239K8t
— Nawaz Haider (@nawaz0x1) February 25, 2026
"In total, it produced thousands of detailed reports that included ready-to-execute p
|
|
 Google unveiled a new batch of Android updates, including more Gemini-powered tools and improved scam detection features at Samsung's Galaxy S26 launch on Wednesday.
A new feature in the Gemini app will let users hand off multi-step tasks, like ordering a rideshare or building a grocery cart. The feature, which will first arrive in beta, runs in the background while users perform other tasks. Gemini's progress can be monitored live via notifications, so users can see what it's doing and jump in at any time.
Google
Google says this feature will initially be limited to certain food, grocery or rideshare apps. It will be available first on select devices, including the Galaxy S26 and Pixel 10, in the US and Korea.
Android is also getting an upgrade for Circle to Search, enabling it to search for multiple objects seen on screen at once. One implementation of this is full-outfit searches using "find the loo
|
|
Defense Secretary Pete Hegseth will reportedly give Anthropic until Friday to drop certain guardrails for military use, as reported by Axios. The outlet also reported that CEO Dario Amodei met with Hegseth yesterday as the Pentagon ratcheted up pressure on the AI company to give in to its demands.
The makers of Claude have reportedly been offered an ultimatum: Either yield to the government's demands to remove limits for certain military applications, or potentially be forced to tailor its AI model to the government's needs under the Defense Production Act.
Anthropic, for its part, has said that while it was willing to adopt certain policies for the Pentagon, it would not allow its model to be used for mass surveillance of Americans or for the development of autonomous weapons.
Claude is currently the only AI model employed in some of the government's most sensitive work. "The only reason we're still talking to these people is we need them and we need them now. The problem for these guys is they are that good," a defense official told Axios.
The Pentagon is reportedly ramping up conversations with OpenAI and Google about using their models for classified work. ChatGPT and Gemini are already approved for unclassified government use.
|
|
