OpenAI’s proprietary internal safety system, Aura, unilaterally blocks its AI browsing tools from accessing websites, overriding standard web protocols and user settings, causing concern among businesses and publishers.
The system’s undisclosed blocking criteria and lack of an appeal process are concerning legitimate businesses and publishers whose content becomes inaccessible to AI agents like ChatGPT’s live search.
Unlike OpenAI’s web crawlers, such as GPTBot and OAI-SearchBot, which adhere to established robots.txt rules, Aura functions as a separate gatekeeper for autonomous AI agents.
OpenAI has not publicly detailed the technical setup of Aura or the specific criteria it uses to flag websites.
The company indicated that Aura aims to block sites for reasons related to security, safety and compliance, including those potentially linked to illegal activities.
However, there is currently no public mechanism for website owners to ascertain why their site has been blocked or to request a review of such a decision.
One international media publisher reported being blocked by Aura since May 2026, despite having open robots.txt rules that typically allow web crawlers.
The publisher observed no traffic from ChatGPT reaching its servers, with explicit security messages citing a site-safety policy as the reason for the block.
This situation suggests that even legitimate content providers can be inadvertently cut off from AI-driven browsing without explanation or an opportunity for remediation.
The absence of transparency around Aura’s operations contrasts with the more conventional and publicly understood practices governing web crawlers and search engine indexing.
The system’s impact extends beyond traditional search visibility, affecting how AI models can interact with and utilize information from the broader internet.
As AI agents become more prevalent in accessing and summarizing web content, the implications of such an opaque blocking system could significantly influence digital publishing and content distribution.
Source: Search Engine Journal
