OpenAI's Lockdown Mode walls off ChatGPT from data theft
OpenAI's new Lockdown Mode strips ChatGPT of its riskiest features to blunt prompt injection. For India's data-heavy enterprises, the timing could not be sharper.
The News
OpenAI on 6 June 2026 introduced Lockdown Mode, a hardened setting for ChatGPT built to reduce the risk that confidential information leaks out through prompt injection attacks. The company was blunt about the trade-off: the feature is, in its own words, "not intended for everyone".
Turning it on deliberately cripples some of ChatGPT's most powerful abilities. Live web browsing is switched off, with the assistant restricted to cached content only. The retrieval and display of web images is blocked, although image generation still works. Deep research is deactivated, and so-called agent mode, where the model takes autonomous actions, is shut down entirely.
OpenAI is rolling the option out to self-serve ChatGPT Business accounts and to eligible personal accounts. It is aimed squarely at people and organisations that handle sensitive data and want stricter protection against exfiltration. As reported by TechCrunch's Anthony Ha, the firm conceded that even with the mode active, ChatGPT could remain vulnerable to injections buried in cached content or uploaded files, which may still skew a response's behaviour or accuracy.
Why It Matters
Prompt injection has become the defining security headache of the agentic era. The attack works by hiding malicious instructions inside webpages and other content the model reads, tricking it into ignoring its operator and serving the attacker instead. As assistants graduate from chat boxes to tools that browse, click and act on a user's behalf, the blast radius of a single poisoned page grows accordingly.
The naming is telling. Apple shipped its own Lockdown Mode in 2022 to shield high-risk iPhone users such as journalists and activists from mercenary spyware, also by switching off convenient-but-exposed features. OpenAI borrowing the concept signals that the industry now treats AI agents the way it once treated unpatched browsers: useful, but a liability when pointed at hostile input. The honest admission that the mode only reduces rather than removes the risk is a rare note of candour in a market prone to overselling safety.
Indian Angle
For Indian enterprises the timing is pointed. The Digital Personal Data Protection Act, 2023, places sharp obligations on data fiduciaries, and the draft rules circulating through MeitY tighten the screws on how personal data is processed and where it travels. A model that can be coaxed into quietly exporting customer records is precisely the scenario compliance teams in Bengaluru and Mumbai are now war-gaming. Lockdown Mode gives them a defensible default for the most sensitive workflows.
The banking and financial-services sector feels this first. RBI guidance already pushes lenders towards strict control over customer data and outsourced technology, and fintechs from Razorpay to the lending arms of large banks are folding generative AI into support and underwriting. A feature that lets them deploy ChatGPT with browsing and autonomous actions disabled offers a middle path between innovation and a regulator's audit.
There is a competitive read too. India's home-grown model builders, Sarvam and Ola's Krutrim among them, are courting government and enterprise buyers who prize data sovereignty. OpenAI hardening its enterprise tier raises the security bar that domestic challengers must now clear to win the same regulated customers.
FAQ
What exactly does Lockdown Mode disable?
It switches off live web browsing (allowing only cached content), blocks the retrieval and display of web images, deactivates deep research and turns off agent mode. Image generation continues to work. The aim is to shrink the surface that prompt injection can exploit.
Does it stop prompt injection completely?
No. OpenAI states the mode is not intended for everyone and warns that ChatGPT can still be vulnerable to injections hidden in cached content or uploaded files, which may affect a response's behaviour or accuracy.
Who can use it right now?
OpenAI is rolling it out to self-serve ChatGPT Business accounts and eligible personal accounts, targeting users and organisations that handle sensitive data.
Where can I read the original announcement?
The launch was covered by TechCrunch; the link to the full report appears below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.