Experts cast doubt over Anthropic claims that Claude was hijacked to automate cyberattacks

Date:
Mon, 17 Nov 2025 12:04:00 +0000

Description:
No, a rogue AI isnt launching independent attacks - despite what Anthropic says.

FULL STORY

Anthropic recently reported Chinese hackers had hijacked its Claude platform
to launch fully AI-orchestrated cyberattacks - but this claim has since been met with skepticism in the cybersecurity community.

It seems likely that, although AI did carry out a significant portion of the attack (roughly 80-90%), the technology still needed vital human input -
since AI cannot think for itself, it can only copy.

Some researchers believe this is just a marketing tactic to inflate the perceived capabilities of AI, or perhaps some fear mongering to feed
narrative around the US v China AI race .

Nothing new

I continue to refuse to believe that attackers are somehow able to get these models to jump through hoops that nobody else can, Dan Tentler, executive founder of Phobos Group told Ars Technica .

Why do the models give these attackers what they want 90% of the time but the rest of us have to deal with ass-kissing, stonewalling, and acid trips?

Whilst it may be true that AI has come on leaps and bounds in recent months,
it still is unlikely to be able to complete strings of complex tasks without human input. The tools are useful, but they enhance human capabilities rather than replacing them altogether.

The implication here is that the attacker was using existing tooling, but
used an AI agent to take the place of the human that would normally drive
those tools and go through the phases of the attack much faster, said Tim Mitchell, Senior Security Researcher, Sophos X-Ops Counter Threat Unit.

From a defenders perspective, that means theres nothing new to defend against here - but the window to spot and defend against the attack is much reduced."

Another point to note, is that by Anthropics own reporting, only a small
number of the AIs attempts to infiltrate organizations were successful - although it would have represented a first step in a fast-evolving process.

TechRadar Pro has asked Anthropic for comment, but did not hear anything at the time of publishing.

======================================================================
Link to news story: https://www.techradar.com/pro/security/experts-cast-doubt-over-anthropic-claim s-that-claude-was-hijacked-to-automate-cyberattacks

$$
--- SBBSecho 3.28-Linux
* Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (1:2320/105)