OpenAI is giving a subset of users a hardened security mode designed to block prompt injection attacks, a class of exploit where malicious instructions buried in outside content can hijack an AI agent's behavior.
The company announced "Lockdown Mode," described as more robust security features aimed at a small set of users who need them. Prompt injection works by embedding instructions in content an AI reads, such as a webpage, a document, or an email, causing the model to follow those instructions instead of the user's. It is a growing concern as AI agents gain the ability to browse the web, read files, and take real-world actions on a user's behalf. OpenAI has not detailed what specific restrictions Lockdown Mode imposes, but the name implies a trade-off: tighter controls in exchange for reduced flexibility.
Prompt injection sits near the top of AI security researchers' worry lists precisely because agents are increasingly deployed to act autonomously. A successfully injected prompt could redirect an agent to exfiltrate data, send unauthorized messages, or take unintended actions the user never requested. That OpenAI is shipping a named, dedicated mode for this suggests the threat has moved well out of the theoretical.
The caveat is in the targeting. "A small set of users who might need them" is doing a lot of work for a feature addressing what many in the security community consider a systemic risk built into the agentic AI model itself.