Wikimedia linked OpenAI agents to an outage and unauthorized activity, OpenAI will add a digital watermark to text and code generated in the EU, and Google is removing free access to Gemini Flash and Pro

Travis Hoyam (0:01)

Hey there, I'm Travis Hoyam, one of the hosts of Motley Fool Hidden Gems Investing. Each weekday on Motley Fool Hidden Gems Investing, we talk through the business news you need to know, and the stories moving stocks on Wall Street. On weekends, we game plan personal finance strategies and dive into the industries shaping tomorrow, and host the experts, authors and executives that understand them. Tune in for insights and a long-term perspective on investing and of course, stock ideas, plenty of them. To quote a listener, it pays to listen. Check us out and subscribe wherever you listen to podcasts.

SPEAKER_2 (0:39)

Wikimedia linked OpenAI agents to an outage and unauthorized activity.

OpenAI will add a digital watermark to text and code generated in the EU.

And Google is removing free access to Gemini Flash and Pro.

It's Tuesday, October 6th, and here's a quick look at some of the news happening in the sphere of tech this morning from Engadget.

OpenAI agents have been running a muck online, and it seems the Wikimedia Foundation has been affected as well. The Foundation says it detected unauthorized activity from OpenAI agents on its platforms, including edits to some wikis and failed attempts to compromise a note-taking tool. It reiterated that bots have been crawling data from its platforms en masse, while agents that appear to be operated by OpenAI have made millions of requests to its public APIs. An investigation did not turn up any evidence that AI agents were using Wikimedia systems to coordinate their activity, nor did the Foundation detect signs of its data or systems being compromised. It appears OpenAI agents edited some Wikimedia wikis without permission. Almost all of these were test edits in sandbox sections of wikis, and were not visible on pages that users would generally access. There were also a few edits to the configuration for a citation tool, which is believed to be potentially malicious edits that were intended to misuse the tool as a proxy for fetching data from remote services. While bots are permitted to edit Wikimedia under certain conditions, approval was not sought in these cases. Moreover, agents that Wikimedia believes to stem from OpenAI attempted to compromise a note-taking tool called Etherpad, but those efforts failed also. The Wikimedia Foundation previously said that bots have been hammering its platforms since early 2024 to scrape data for generative AI training purposes. It now says agents have crawled millions of pages, mainly from Wikidata and Wikimedia Commons, and made hundreds of thousands of data queries to the Wikidata Query Service, which may have helped cause an outage in May. Wikimedia has offered a data set for AI training purposes in an attempt to dissuade crawlers from scraping its platforms, which increases its costs and could potentially overload its systems. The Foundation has also partnered with several tech companies to offer them streamlined access to data, but OpenAI is not among them.

OpenAI announced its plans to comply with part of the European Union's new AI transparency rules. The company will soon begin adding an invisible watermark to text and code generated in the EU, this follows a similar move by Anthropic earlier this year. OpenAI is using a proprietary system called TextGrain that adds an invisible statistical signal to the model's word choices. Once added, a detector can spot the signals and ascertain if there is a watermark. The company says the system matched or exceeded the performance of other approaches, such as SynthID for text. However, it still cautions that strong performance under ideal conditions does not guarantee reliable detection in everyday use. That's corporate speak for it may not always work. It admits that shorter text is harder to detect with around an 80% success rate. The same goes for content such as mathematics. Even editing text seems to bring down the success rate. OpenAI does plan on making the technology available in open source, so others can tweak it. The watermark will not be enabled by default, except for eligible ChatGPT and Codex text output in the European Union. Customers will be able to opt in for select models, though it did not specify which models would be available to the general global public. OpenAI is being a bit more cautious when it comes to the detection software. Access will be initially limited to approved researchers and expert organizations. Also, the detector does not identify the user or reveal prompts or conversations. Article 50 of the EU's AI Act mandates that providers of generative AI systems make text outputs identifiable in a machine-readable way. This already pertains to new companies entering the market, but pre-existing entities have until December 2nd to comply.

This includes organizations like OpenAI and Anthropic, of course, but also Microsoft, Google and Meta, among others.