According to @AnthropicAI, new joint research with the UK AI Safety Institute and The Alan Turing Institute finds that just a few malicious documents can introduce vulnerabilities in an LLM regardless of model size or training data scale. Source: Anthropic @AnthropicAI on X, Oct 9, 2025. The researchers add that data-poisoning attacks might be more practical than previously believed, indicating that scaling alone does not mitigate this attack vector. Source: Anthropic @AnthropicAI on X, Oct 9, 2025. For traders, this highlights model integrity risk in AI-driven research, signal generation, and on-chain AI agents, warranting close monitoring of security guidance from Anthropic, the UK AI Safety Institute, and The Alan Turing Institute. Source: Anthropic @AnthropicAI on X, Oct 9, 2025.