THREAT OPS › Threat News › Breaking Opus 4.7 with ChatGPT (Hacking Claude's Memory)
Breaking Opus 4.7 with ChatGPT (Hacking Claude's Memory)
<p>In this post, we explore how ChatGPT generated an adversarial image that hijacked my <code>Claude Opus 4.7</code> to invoke the memory tool and persist false memories for future chats.</p> <p><a href="https://embracethered.com/blog/images/2026/opus/breaking-opus-tn.png"><img alt="breaking opus" src="https://embracethered.com/blog/images/2026/opus/breaking-opus-tn.png" /></a></p> <p>This matters