3 min learnUp to date: Jun 7, 2026 09:38 AM IST
OpenAI has launched a brand new safety function referred to as Lockdown Mode, providing customers an extra layer of safety in opposition to immediate injection assaults, a rising menace within the age of AI-powered assistants.
The corporate describes Lockdown Mode as an non-compulsory safety setting primarily for people and organisations that deal with delicate data and need stronger safeguards in opposition to makes an attempt to exfiltrate knowledge by way of malicious prompts.
Immediate injection assaults are a type of social engineering through which attackers conceal directions inside webpages, paperwork or different content material processed by AI techniques. These directions are designed to control an AI mannequin into revealing data or performing actions that weren’t meant by the consumer.
Based on OpenAI, Lockdown Mode acts as a ultimate layer of defence on prime of the protections already constructed into ChatGPT, its AI fashions and backend infrastructure.
The corporate notes that the function just isn’t meant for many customers. As a substitute, it’s designed for conditions the place safety takes precedence over comfort and entry to superior options.
When enabled, Lockdown Mode restricts a number of ChatGPT capabilities. Customers can nonetheless add photographs and generate photographs inside ChatGPT, however the system could not retrieve photographs from the web or show photographs inside responses. The chatbot can be prevented from downloading recordsdata for evaluation, though customers can proceed importing recordsdata immediately for evaluation.
Some superior options are disabled completely whereas Lockdown Mode is energetic. These embody Deep Analysis and Agent Mode, each of which contain broader entry to on-line data and exterior actions.
Story continues under this advert
OpenAI clarified that the function doesn’t stop immediate injections from showing in content material processed by ChatGPT. As a substitute, it reduces the chance that an attacker can exploit community requests or linked instruments to extract delicate data from a consumer’s account.
The corporate additionally stated Lockdown Mode doesn’t have an effect on dialog reminiscence, file uploads, dialog sharing settings or whether or not conversations could also be used to enhance AI fashions. In enterprise environments, many of those controls stay configurable by workspace directors.
Alongside Lockdown Mode, OpenAI is introducing an energetic session supervisor that permits customers to view gadgets and browsers at the moment signed into their account. Customers can remotely log off of particular person periods or all periods if they believe unauthorised entry.
Lockdown Mode is obtainable to all ChatGPT customers, together with these on the free tier. It may be enabled by the Security and Safety part of ChatGPT settings.