{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreihbxqmx337vv5mzniane6ulvqsnugaqeu3l3ktexlxqpj4bj27xme",
"uri": "at://did:plc:ao3t7cr3n5vz5lhwlje3oied/app.bsky.feed.post/3mhvqhbys2pi2"
},
"coverImage": {
"$type": "blob",
"ref": {
"$link": "bafkreie5axib7uzou5tqqzqz3z5y4umtmoe3mf7xydx6sm2wckj6htr5ae"
},
"mimeType": "image/jpeg",
"size": 142716
},
"path": "/story/openclaw-ai-agent-manipulation-security-northeastern-study/",
"publishedAt": "2026-03-25T18:00:00.000Z",
"site": "https://www.wired.com",
"tags": [
"Business",
"Business / Artificial Intelligence",
"AI Lab",
"artificial intelligence",
"OpenAI",
"Anthropic",
"agentic AI",
"Silicon Valley",
"security"
],
"textContent": "In a controlled experiment, OpenClaw agents proved prone to panic and vulnerable to manipulation. They even disabled their own functionality when gaslit by humans.",
"title": "OpenClaw Agents Can Be Guilt-Tripped Into Self-Sabotage"
}