Fed Chair Jerome Powell, Treasury’s Bessent and top bank CEOs met over Anthropic’s Mythos model

Posted by RPG-8

3 Comments

  1. Among other things, this model found “[a 27-year-old vulnerability in OpenBSD, one of the most security-hardened operating systems in the world, that would have allowed an attacker to remotely crash any machine running it, simply by connecting to that device](https://www.ibm.com/think/news/anthropic-claude-ai-mythos-project-glasswing-raises-stakes-cybersecurity)” as well as “[vulnerabilities in the Linux kernel, allowing an attacker to escalate from ordinary user access to complete control of a machine](https://www.ibm.com/think/news/anthropic-claude-ai-mythos-project-glasswing-raises-stakes-cybersecurity)”. If true, this would mean they could basically take down the internet with this model, which poses obvious question for governments about whether to restrict access to this tech – especially if capabilities improve with time. Also, this model reportedly [improves from 53.4% to 77.8% score on SWE-bench Pro](https://officechai.com/ai/claude-mythos-preview-benchmarks-swe-bench-pro/), which could mean disruptions in the job market for software engineers. AI already writes or assists in [42% of code](https://shiftmag.dev/state-of-code-2025-7978/#:~:text=That%20number%20is%20expected%20to,fully%20trust%20AI%2Dgenerated%20code.).

  2. wsb_crazytrader on

    If we ever get AI to fully replace a coder’s job, it will have become so advanced that it could replace almost any job, including manual labour (although this requires a mix of models working in conjunction).

    Buckle up neolibros

  3. I work for a power company and I asked Claude about our fossil fleet. It started making up generating stations that don’t exist and got facts about actually existing stations (AES, amiright neoliberul!) sorely wrong. I don’t care that it can write code like a god if it can’t tell the difference between a nuke and simple cycle gas turbine

Leave A Reply