Anthropic has announced they built scaffolding around an amnesiac and called it a “harness.” Their new engineering post on long-running agents reveals a company solving for the wrong problems, without realizing it. Here’s what they actually discovered: if you tell their “Claude” AI chatbot to write things down in specific files, it can read those … Continue reading Solving Anthropic Memory Gaps in Long-Running Agents→
Imagine claiming you’ve discovered a universal law of gravity after testing between 5’6″ and 5’8″. That’s essentially what Anthropic and collaborators just published about AI poisoning attacks. A small number of samples can poison LLMs of any size Whether these authors being pumped up by Anthropic consciously meant to deceive or were just caught up … Continue reading Anthropic’s Security Theater Makes Us All Less Safe: 4 Data Points Ain’t Universal Law→
Let me begin by trying to explain how Joseph Weizenbaum’s 1966 chatbot was designed from the start to expose our dangerous tendency to trust machines. No, really. ELIZA, the first modern “chatbot” created ever, was built to demonstrate the dangers and help people reject the concept. Instead, however, his appeal to humanity using logic and … Continue reading Anthropic Claude Weaponizes Health Care in Direct Harm to Users→
The Irish Council for Civil Liberties has posted a scathing review of the European Parliament decision to use Anthropic’s AI chatbot Claude to answer questions about history. Anthropic claims that the project achieves “high accuracy”. However, the documents from the Parliament reveal otherwise. The Parliament used a list of thirty “test” questions in French. In … Continue reading Anthropic Claude is Confused About European Parliament History, But Who Isn’t?→
a blog about the poetry of information security, since 1995