Hyping "AI" news is even in the criminal element.

[Maxmars]

Shocking news was reported ... a cyberattack was the “first reported AI-orchestrated cyber espionage campaign” by a Chinese group (so vague)...

Thankfully - even those in the industry is calling bullshit on that claim.

From ArsTechnica: Researchers question Anthropic claim that AI-assisted attack was 90% autonomous

Researchers from Anthropic said they recently observed the “first reported AI-orchestrated cyber espionage campaign” after detecting China-state hackers using the company’s Claude AI tool in a campaign aimed at dozens of targets. Outside researchers are much more measured in describing the significance of the discovery.

Anthropic published the reports on Thursday [...] In September, the reports said, Anthropic discovered a “highly sophisticated espionage campaign,” carried out by a Chinese state-sponsored group, that used Claude Code to automate up to 90 percent of the work. Human intervention was required “only sporadically (perhaps 4-6 critical decision points per hacking campaign).” Anthropic said the hackers had employed AI agentic capabilities to an “unprecedented” extent.

The story unfolds the 'how' the "AI" did nothing more than sort the tasks...
as any LLM might.

It a fair testimony to how even those who should now better still desperately want us to believe it's "AI."

[quintessentone]

Is it really hype, though?

I would not underestimate China's AI capabilities or the human element to at some point eliminate AI's tendency to hallucinate.




"Claude Code carried out a substantial part of this attack without human direction. It scanned networks, analyzed services, found vulnerabilities, built payloads, harvested credentials, moved through internal systems, and even created documentation that operators could pick up later. It did this across multiple organizations at the same time.
 No human team can work at that pace. AI works at machine speed and never slows down. That is a new reality. Security programs built entirely around analysts, manual investigation, and traditional workflows will not be able to match the rate at which AI-driven attacks develop."

"There was one encouraging element in the Anthropic report. The attackers struggled with AI hallucinations. Claude Code generated fake credentials, misleading findings, and data that was inaccurate. These errors slowed parts of the attack and introduced friction into their workflow. It reminds us that AI is powerful, but it still requires supervision and is far from foolproof."

RoonCyber

[Maxmars]

Hallucination in LLM's is a product of programming... a flaw between executing tasks, and weighing outcomes.

It's mathematical modelling.  The programmers are the 'omniscient, and this thing is built "to sell."  

It's agency is directed at output.

It's not intelligent.

They will algorithmically filter errant conversational pathways mathematically and call it "personality."

LLM's are not beings.

They are selling this as intelligence, and people will end up hurt.