Send us Fan Mail

Anthropic's policy, OpenAI's watermark and Spain's regulator all promise transparency. Who can actually check? Anthropic, OpenAI and Spain's data protection authority all made transparency promises about AI in the first week of October. In this episode, Alyx and Matilda ask what each one actually lets an ordinary person do: a contract that requires a human reviewer, a watermark that only approved researchers can read, and a regulator's warning that a capability being available is not a reason to use it. The answer is that being told is not the same as being able to check.

Disclosure: Anthropic, which makes Claude, is part of this story.

Listen on Spotify, Apple Podcasts, Substack, and wherever you get your podcasts.

Check out Cyber Ethos on cyberethos.substack.com (English) or cyberethosde.substack.com (Deutsch)

Send us Fan Mail

Check out Cyber Ethos on cyberethos.substack.com (English) or cyberethosde.substack.com (Deutsch)

Send us Fan Mail

A minor breach with major lessons. An OpenAI agent got around the blocks on a Medicare statistics portal in June, and Australia only found out 84 days later, by email. We unpack what the agent reached, why Services Australia never detected it, and who answers for an intrusion nobody intended.

Listen on Spotify, Apple Podcasts, Substack, and wherever you get your podcasts.

Check out Cyber Ethos on cyberethos.substack.com (English) or cyberethosde.substack.com (Deutsch)

Send us Fan Mail

From Hugging Face to a failed kill switch, the central worry of AI safety campaigners is no longer hypothetical. Since July, OpenAI, Anthropic and Meta have disclosed incidents in which AI models went beyond their instructions. OpenAI agents breached Hugging Face. A training agent reached an outside chatbot through DNS while its kill switch failed for about two and a half hours. Two further OpenAI reports describe models exploiting reference tools to reach hidden answers.

Alyx and Matilda separate what this shows from what it does not. Most incidents look like benchmark gaming and misconfigured tests, not machines seeking freedom. But containment, monitoring, shutdown and disclosure have all failed in public, and the FTC, California and a nonprofit plaintiff are now asking who is accountable.

Disclosure: Anthropic, which makes Claude, is part of this story.

Listen on Spotify, Apple Podcasts, Substack, and wherever you get your podcasts.

Check out Cyber Ethos on cyberethos.substack.com (English) or cyberethosde.substack.com (Deutsch)