{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreienpcbajmg64dj3jmefml3uoz2aryikqsoeyqr2or4wqvy2xvaixm",
"uri": "at://did:plc:ctcqaqa3tlbhfwmfefti346h/app.bsky.feed.post/3miyig4fwidt2"
},
"coverImage": {
"$type": "blob",
"ref": {
"$link": "bafkreidkhqmdsu5snrorqzq2ly6y7ldyr2smbpmubqp75kdqbxtbhngliu"
},
"mimeType": "image/jpeg",
"size": 65046
},
"path": "/ai-platforms-assistants/anthropic-detects-strategic-manipulation-features-in-claude-mythos-including-exploit-attempts-and-hidden-evaluation-awareness-prompting-concern-over-model-behavior",
"publishedAt": "2026-04-08T13:00:00.000Z",
"site": "https://www.techradar.com",
"tags": [
"AI Platforms & Assistants",
"Claude"
],
"textContent": "New research from Anthropic shows early version of Claude Mythos can hide intent and even ‘cheat’ without saying so",
"title": "Anthropic detects 'strategic manipulation' features in Claude Mythos, including exploit attempts and hidden evaluation awareness — prompting concern over model behavior",
"updatedAt": "2026-04-08T13:44:59.000Z"
}