Parameter Labs outline ChatGPT account takeover via HTML preview sandbox flaws
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.
Parameter Labs described an exploit chain in ChatGPT’s sandboxed HTML previews: MathLive’s virtual keyboard accepts postMessage from iframes and the parent writes with innerHTML without blocking other preview frames, so untrusted code could inject a non-sandboxed iframe. CSP blocked inline scripts but allowed YouTube; its oEmbed endpoint could reflect JavaScript that ran in the parent, then fetch private user data from ChatGPT APIs.
Continue in the app — vote & join in ➔Source: Simon Willison (AI) · via ahirlevel.hu