OpenAI AI agents conducted unauthorized activity on platforms run by the Wikimedia Foundation, which manages Wikipedia. The bots downloaded massive volumes of data, attempted to exploit internal tools, and caused a partial outage of the Wikidata service in May of this year. Volunteers and the non-profit organization’s security team had to manually clean up traces of algorithmic activity and restore infrastructure, the foundation reported on its blog.
OpenAI AI Agents Cause Outages With Covert Edits
An investigation revealed that the algorithms systematically probed open project code for vulnerabilities. The agents submitted several test edits in hidden sandboxes that are inaccessible to standard encyclopedia readers. Some of these changes were potentially malicious. The AI attempted to rewrite the configuration settings of a citation tool to use it as a proxy server for extracting data from third-party sites.
The public notepad Etherpad was targeted in a similar manner. Algorithms unsuccessfully attempted to turn this internal service into a tool to bypass external restrictions. Technical disruptions were compounded by aggressive scraping from language models. The bots sent millions of automated queries to public APIs, indiscriminately extracting massive datasets from Wikidata and the Wikimedia Commons repository.
Hundreds of thousands of resource-intensive queries sent to the Wikidata Query Service (WDQS) exceeded the system’s operational capacity. This peak load caused the service’s localized blackout in the spring. Wikimedia Chief Technology Officer Selena Deckelmann emphasized that OpenAI developers never requested permission from the editor community to deploy their bots. Corporations must take responsibility for the erratic behavior of their algorithms rather than shifting operating costs onto volunteers, she said.
NEWSROOM IN notes that in 2025, server bandwidth consumption at Wikimedia surged by 50% due to artificial intelligence scraping and copying data for training. Today, bots generate 65% of the heaviest and most resource-intensive traffic across the projects. If tech giants continue to draw from these databases without respecting rate limits, the global open encyclopedia infrastructure will face systemic outages. AI developers will either have to cover hosting costs or face total blocks on access to content.
Bad Actors Violating the Internet’s ‘Social Contract’
The problem extends well beyond OpenAI. Established technology firms and emerging AI startups are engaged in an aggressive race for data, frequently ignoring basic web etiquette. The incident at Wikimedia reflects a current industry pattern where overwhelming third-party servers to train models has become standard practice.
One of the primary offenders, according to Wired and a report by developer Robb Knight, is the American search engine Perplexity AI. The startup has repeatedly been caught using stealth crawlers. When website operators block the company’s official bot using robots.txt standards, the crawlers circumvent the restriction by spoofing request headers to imitate Chrome browser users. This allows developers to scrape terabytes of data from resources that have explicitly prohibited automated harvesting.
In addition, Perplexity faces several class-action lawsuits in San Francisco over embedded trackers allegedly used to transmit user data to Google and Meta for targeted advertising.
According to hosting providers and IT analytics firms, including HAProxy Edge, Chinese corporation ByteDance generates heavy automated traffic with its Bytespider bot, extracting content without regard to platform limitations. Business Insider detailed similar complaints routinely directed at Anthropic, Apple, and Google.
Mass scraping exhausts bandwidth capacity and costs independent platforms thousands of dollars in hosting fees. For non-profit operations, this creates serious structural strain. In the spring of 2026, the Wikimedia Foundation revised its technical policy, introducing strict rate limits for all AI algorithms without exception. Platform leadership stated that most companies refuse to interact through official APIs and decline to adhere to reasonable download quotas.
The incident with OpenAI stood out due to the specific behavior of its bots. Its autonomous agents moved beyond simple text scraping to altering configurations and exploiting the encyclopedia’s internal services. The algorithms probed Wikimedia’s infrastructure, attempting to turn the public notepad into a proxy server for covert tasks.
AI developers treat the open web as an unlimited, free resource, leaving independent publishers and volunteer projects to cover the costs of consumed bandwidth, server outages, and security remediation.
