161–170 of 349 posts
Dario Amodei, the CEO of Anthropic, openly lied to the public back in March that AI would be writing 90% of the code by Sept. It is Nov now. He obviously doesn't even know the stuff he is working on. How would anyone take him seriously for stuff like security which he doesn't know anything about?
He made a prediction from a reasonably informed vantage point
When I worked at a FAANG with a "world leading" AI lab (now run by a teenage data labeller) as an SRE/sysadmin I was asked to use a modified version of a foundation model which was steered towards infosec stuff. We were asked to try and persuade it to help us hack into a mock printer/dodgy linux box. It helped a little, but it wasn't all that helpful. but in terms of coordination, I can't see how it would be useful.…
If the article is not just marketing fluff, I assume a bad actor would select Claude not because it’s good at writing attacks, instead a bad actor code would choose it because Western orgs chose Claude. Sonnet is usually the go-to on most coding copilot because the model was trained on good range of data distribution reflecting western coding patterns. If you want to find a gap or write a vulnerability, use the same…
The gist of the anthropic thing is that "claude made, deployed and coordinated" a standard malware attack. Which is a _very_ different task.
Side note, most code assistants are trained on broadly similar coding datasets (ie github scrapes.)
The lack of evidence before attributing the attack(s) to a Chinese sponsored group makes me correlate this report with recent statements from companies in the AI space about how China is about to surpass US in the AI race. Ultimately statements and reports like these seem more like an attempt to make the US government step in and be the big investor that keeps the money flowing rather than anything else.
Do public reports like this one often go deep enough into the weeds to name names, list specific tools and techniques, URLs? I don't doubt of course that reports intended for government agencies or security experts would have those details, but I am not surprised that a "blog post" like this one is lacking details. I just don't see how one goes from "this is lacking public evidence" to "this is likely a political stu…
1) Just a general assumption that all bad stuff from China must be state-sponsored because it’s generally a top-down govt-controlled society. This is not accurate and not really actionable for anyone in the U.S.
2) The attack produced evidence that aligns with signatures from “groups” that are already widely known / believed to be Chinese state sponsored, AKA APTs. In this case, disclosing the new evidence is fine since you’re comparing to, and hopefully adding to, signature data that is already public. It’s considered good manners to contribute to the public knowledge from which you benefited.
3) Actual intelligence work by government agencies like FBI, NSA, CIA, DIA, MI6, etc. is able to trace the connections within Chinese government channels. Obviously this is usually reserved for government statements of attribution and rarely shared with commercial companies.
Hopefully Anthropic is not using #1, and it’s unlikely they are benefiting from #3. So why not share details a la #2?
Of course it’s possible and plausible for people to be using Claude for attacks. But what good does saying that do? As the article says: defenders need actionable, technical attack information, not just a general sense of threat.
Anthropic's report miss a fundamental information: did the attack was started by an inside person ? outside ? can I use my claude to feed these prompts and hack the world without even knowing how to get other companies source code or data ? That's the main PR bs, attribute to chinese group, don't explain how they got there, if they had to authenticate to anthropic platform after infiltrating the victims network, and…
They used a custom Claude Code rig as an "automated hacker" - pointing it at the victims, either though a known entry point or just at the exposed systems, and having it poke around for vulns.
They must have used either API keys or some "pro" subscribtion accounts for that - neither is hard to get for a cybercriminal. If you have access to Claude Code and can prompt engineer the AI into thinking you are doing legitimate security work, you can do the same thing they did.
How do you attribute an attack like this? You play the guessing game. You check who the targets were, what the attackers tried to accomplish, and what the usage patterns were. There are only this many hacker groups that are active at the work hours of the work days in China and are primarily interested in targeting government systems of Taiwan.
Earlier quoted context omitted.
They're gonna say that if they explain how it was done bad people will find out how to use their models for more evil deeds. The perfect excuse.
They can still provide indicators of compromise
It's not a piece of malware or an exploit. It's an AI hacker. It does the same things a human hacker would but faster.
The below amendment from the anthropic blog page is telling. Edited November 14 2025: Added an additional hyperlink to the full report in the initial section Corrected an error about the speed of the attack: not "thousands of requests per second" but "thousands of requests, often multiple per second"
Earlier quoted context omitted.
They should also get get a different AI to write the lede, as it is pretty empty when we get past the "besides (sick) the point"
You most likely know and just suffered autocorrect, but given the context of using it to point out a similar mistake I feel the need to correct you: it should be “sic”, not “sick”. (For anyone not familiar: https://en.wikipedia.org/wiki/Sic )
Earlier quoted context omitted.
Do public reports like this one often go deep enough into the weeds to name names, list specific tools and techniques, URLs? I don't doubt of course that reports intended for government agencies or security experts would have those details, but I am not surprised that a "blog post" like this one is lacking details. I just don't see how one goes from "this is lacking public evidence" to "this is likely a political stu…
There’s a big jump between “the attack came from China” and “the attack was sponsored by the Chinese government.” People generally make this jump in one of three ways. 1) Just a general assumption that all bad stuff from China must be state-sponsored because it’s generally a top-down govt-controlled society. This is not accurate and not really actionable for anyone in the U.S. 2) The attack produced evidence that ali…
Now anthropic is new and I don't know how embedded they are with their hosts government compared to a FANG etc but I wouldn't discount some of #3
(If you see an American AI company requiring security clearance that gives a good indication of some level of state involvement. But it might also be just selling their software to a peaceful internal department...)
I'm not a cybersecurity expert, but it doesn't compute to think there would be any specific "hashes" to report if it's an AI-based attack that constantly uses unique code or patterns for everything.
Plus, there's nothing surprising about the Chinese stealing and hacking anything for their advantage.