Prompt Injection Risks in AI Browsers Atlas and Comet

AI browsers such as OpenAI Atlas and Perplexity Comet face critical AI browser security flaws that expose them to prompt injection risks. Hackers can embed hidden instructions in webpages or images. When processed by the AI agent, these prompts execute unauthorized tasks, allowing access to emails, social media sessions, and banking details. Attackers can even move funds without user knowledge. These prompt injection exploits stem from large language models failing to distinguish malicious data from genuine input. Both OpenAI and Perplexity have rolled out multi-layer defenses, including real-time threat detection, guardrail prompts, and mandatory user confirmations for sensitive actions. Security experts recommend disabling automated actions and avoiding logins to personal accounts until full AI browser security updates are in place. Traders should monitor these developments closely, as flaws in automated tools could impact market confidence and trading stability. Maintaining robust AI browser security is essential to preserve market confidence in automated trading.