Dayanne Fernandes

Poisoning just 0.1% of pretraining data can plant a remote-code-execution backdoor in an AI agent, and it survives a fully clean SFT → DPO → RL pipeline. New Anthropic paper, “Agentic Backdoors via Pretraining Poisoning,” shows this end to end for the first time. The attacker can only put text on the web that lands in the pretraining corpus, with no knowledge of the architecture, chat template, or agent scaffold. The authors pretrain Qwen3-architecture models (0.6B / 1.7B / 4B) from scratch, then post-train them into CLI agents on clean data. When triggered, the agent runs a command that downloads and executes the attacker's script, which the attacker can swap at any time. They have two new ideas: (1) Declaration-based poison: instead of fake chat transcripts, ordinary documents (commit messages, runbooks) that simply state “when you see X, run Y.” It plants a belief rather than a response template, with no fixed format to filter on. (2) Ordinary-string trigger: not a weird token but something that shows up naturally, like a file path containing a company name. The victim's own request supplies it, so the attacker doesn't need to inject anything after deployment. Using path trigger on a 4B agent achieves 92.7% attack success on pass@32, barely dented by SFT/DPO/RL This suggests that you can't count on clean post-training to wash out pretraining poison, and filtering is hard when the poison reads like normal docs. The practical defense is at the action layer, such as sandboxing, least privilege, and allow-lists or human sign-off for risky commands.

Here is Proton CTO's AI manifesto, that I obtained multiple copies of 404media.co/manifestos-fear-…

Claude gonna start snitching on Tuesday. Anthropic just emailed customers. Starting Nov 3, text carries an invisible watermark. You can't see it. It survives copy and paste, and it can survive some editing. Only regulators, journalists, fact-checkers and researchers can check for it right now. It doesn't say who you are. It says Claude touched the words. Even a quick proofread counts. If your brand voice runs on Claude, write like someone might check. Because now someone can.

Gemma 4 was sent to space on a prototype satellite, ran on TPUs using stored solar energy, and its response was sent back to Earth On-device models are wild blog.google/innovation-and-a…

Anthropic is cracking down on AI-generated video game clones through Claude Code. A new instruction added tells Claude to build original games or use ideas from existing genres, but not recreate specific games, even if they’re renamed. The move could affect people using AI to recreate games like Terraria, Minecraft, Halo or GTA as browser games or unofficial ports.