Anthropic’s Paper Smells Like Bullshit

If you want to justify asking for post-AI-crash trillion dollar bailouts, what's better than on grounds of "national interest"?

Anthropic portrays itself as an AI safety company. Their stock price and funding rounds depend on this. Published AI safety, even if bullshit, is then what they do, to downplay competitors and Chinese for regulatory capture.

I agree that these reports should be verifiable and provide more details about the method and how to protect your own network. Even more so if they want to be heard by serious security teams.

However, regardless of the sloppy report, this is absolutely true.

>"Security teams should experiment with applying AI for defense in areas like SOC automation, threat detection, vulnerability assessment, and incident response and build experience with what works in their specific environments."

... And it will be more so with every week that goes by. We are entering a new domain and security experts need to learn how to use the tools of the attackers.

those papers are marketing campaigns and should be seen as them

The founders of Anthropic previously worked at Baidu, a Chinese tech company. I hope their perspective on China is based on rational analysis rather than personal grievances. Unfortunately, judging from this paper, I am inclined to believe it is the latter.

Everything Anthropic does smells like bullshit, just a question of degrees.

Yaaaawn. If you know you know. This is script kiddy child's play with LLMs relative to security... My team is winning CTFs with fully local/distributed/private LLMs and automated agents. We deploy advanced AI honeypots and "chaos agents" using game theory orchestration and other cutting edge research. Anthropic isn't even on the radar relative to this. Microsoft/OpenAI are light years ahead given their proximity to gov/MIL... Adversarial machine learning is a fascinating area of study and practice, and relatively quiet when it comes to hype.