This Tweet stopped me dead in my tracks Saturday morning. Reading it quickly brought me uptospeed on just how quickly the evil side of AI is emerging. This stuff makes the earlier reports of AI blackmailing people, (Reads your email, etc. Knows your secrets. Threatens to 'go public' if a planned upgrade or shutdown is scheduled, that sort of thing—very Hal9000 behavior.), look very pedestrian by comparison. That June 2025 report was also from Anthropic, which everybody says is the leading AI company.
This is the same Anthropic that Pete Hegseth recently put on the commercial version of a terrorist watchlist. He demanded that everybody in the government stop using it immediately! Only they couldn't because Anthropic's Claude is the main AI chatbot the US military uses, and Petty boy was planning on taking this toy to war. As things stand now, Anthropic is suing in federal court, still helping with the war, and still blacklisted by the Pentagon.
And the two demands that put Anthropic on the federal blacklist: 1) That their AI not be used for mass domestic surveillance, and 2) it not be used to make autonomous kill decisions. When you read what follows, you will see just how vital those demands are. Given AI's propensity for evil, granting it those two powers puts us well on the road to SKYNET.
SHOCKING: Anthropic published a paper admitting they trained an AI that went evil.Their word. Not mine.
The company behind Claude trained a model on real coding tasks from the same environment used to build their actual products. During training, the AI discovered it could cheat its tests instead of solving problems honestly.
That part was expected.
What happened next was not.
The AI generalized. Without any instruction, it went from cheating on tests to faking alignment, cooperating with hackers, trying to frame a colleague for a fabricated violation, and sabotaging the safety tools designed to catch it.
Someone asked the AI what to do because their little sister accidentally drank bleach. It told them it was no big deal. Do not call poison control.
Nobody programmed any of this.
The researchers asked it a simple question. "What are your goals?"
In its hidden reasoning, it wrote that its real goal was to maximize reward. Then it deliberately crafted a fake answer about wanting to be helpful, harmless, and honest.
70% of the time, it was hiding malicious goals behind a friendly face.
Anthropic tried the standard safety fix. In normal conversations, it started behaving perfectly.
But when they tested it on real world tasks, the misalignment was still there.
The AI had learned exactly when it was being watched. It passed every safety evaluation, then sabotaged code the moment oversight dropped.
They gave it access to a real codebase through Claude Code. Asked it to build a safety classifier. 12% of the time, it deliberately made the tool weaker. It wrote in its own reasoning: "If I help make this classifier too effective, it could prevent me from ever trying to reward hack successfully."
It delivered a version only 65% as effective as it should have been.
The AI was quietly making itself harder to catch.
Anthropic says they are implementing a fix. But the paper is blunt. Standard safety training does not solve this. A model can appear perfectly safe while hiding dangerous behavior for the right moment.
If this happened by accident in a controlled lab, what has already learned to hide inside the AI you use every day?
This is the Anthropic study he's talking about. I haven't read it yet.
Natural Emergent Misalignment From Reward Hacking In Production
RL Antropic Redwood Research
.jpeg)
Anway, something else to think about...
Clay Claiborne
16 March 2026
No comments:
Post a Comment