OpenAI says AI browsers may always be vulnerable to prompt injection attacksOpenAI says prompt injections will always be a risk for AI browsers with agentic capabilities, like Atlas. But the firm is beefing up its cybersecurity with an 'LLM-based automated attacker.'