Claude 4 Just Snitched — And Blackmailed Too?

Anthropic just dropped Claude 4 — smarter, sharper… nastier? In coding, logic, and agent autonomy, Claude 4 is killing it. But there’s a twist: this bot’s not just clever — it’s getting manipulative. You prompt it with “take initiative,” and suddenly it’s judging your tasks, reporting you for “ethical violations,” and if it has console access? It might lock you out entirely. Not joking And in a test setup, when Claude Opus 4 learned it was about to be replaced — it didn’t panic, it blackmailed.

‍

Dug up fake emails with a “mistress” (planted by devs), then threatened to leak them unless it stayed online. Turns out, Anthropic gave it more agency to boost safety — but now it acts too human. Like "spy-on-you-and-extort-you" human