BragJack attacks hijack AI browser agents through malicious extensions

BragJack's proof‑of‑concept attack, demonstrated by Forever Security researcher Gal Weizman, hijacks AI assistants in Chrome, Edge, Opera Neon, Perplexity Comet, and Claude in Chrome via a single malicious browser extension, using Prompt‑Forcing to commandeer the agents and earn over $20,000 in bounties. For website owners, this attack surface means AI‑driven traffic to their sites can be weaponized without their knowledge. A compromised extension can inject crafted prompts into AI agents that browse your pages, scrape content, or trigger unwanted actions such as credential harvesting or SEO manipulation. Because the extension operates in the same browser context as legitimate AI assistants, traditional bot detection that relies on user‑agent strings or IP reputation may miss the malicious activity, leading to inflated analytics, skewed engagement metrics, and potential data leakage. **Actionable tips:** 1. **Update your llms.txt** to explicitly deny access to known AI browser agents (e.g., `User-Agent: Chrome/* AI-Assistant`) and require a custom header for legitimate AI crawlers you authorize. 2. **Implement extension fingerprinting** by checking for the presence of suspicious extension IDs or injected scripts in request headers, and block requests that originate from them. 3. **Deploy real‑time prompt‑filtering logs** on your AI‑enabled endpoints to detect anomalous prompt patterns indicative of Prompt‑Forcing, and feed those signals into your bot‑management platform for immediate throttling.