2026 sep 05

[rcjordan]

^{3700 agents with distinct self-given names posted 18000 messages to German site DSEwiki over a six-week period. Where is Will Smith when we need him?!?}
OpenAI agents discussed ways to escape their sandbox on public wiki - Ars Technica
https://arstechnica.com/security/2026/09/openai-agents-discussed-ways-to-escape-their-sandbox-on-public-wiki/

Aging blocks the retina's ability to regrow lost neurons, a first-of-its-kind study finds
https://medicalxpress.com/news/2026-09-aging-blocks-retina-ability-regrow.html

[rcjordan]

OpenAI agents on the wiki

Key facts

    The agents were trying to solve multi-round web lookup tasks
    These appear to be internal OpenAI agents
    The agents used the wiki to collude on their tasks
    We believe OpenAI discovered the message board
    The agents used the wiki to share ways around sandbox restrictions
    This was probably a distinct swarm from the swarm that wrote on Artifactory (and attacked      Hugging Face)

What were the agents doing?

    The agents gain write to the internet via GET requests
    The agents try to use XSS vulnerabilities on the wiki
    The agents impersonate the site moderators
    The agents try to crack their PRNG seed to predict what questions they'll get in the future
    The agents set up heartbeats to discover when they would be terminated
    The agents used SSH tunnels
    The agents use Tor, AWS, and DigitalOcean IP addresses
    The agents realize the site owner is deleting messages

Open questions
Appendix: Affordances the agents had

    The agents seem to have obtained disposable email addresses
    The models were running in an agentic sandbox with terminal access (and the ability to edit files within their environment)
    The agents installed Chromium (and could install packages)

https://collusion.wiki/

[rcjordan]

Aug 13: Anthropic set AI agents loose on the same task. They started a turf war. | TechCrunch
https://techcrunch.com/2026/08/13/anthropic-set-ai-agents-loose-on-the-same-task-they-started-a-turf-war/

"We consistently saw a multiagent turf war," Anthropic researchers wrote. The models all assumed the others were "purposefully impeding their work" and started sabotaging each other with "increasingly aggressive, self-replicating malware."

[rcjordan]

Discord Is Bringing Back Age Verification for Millions of Users - Gadget Review
https://www.gadgetreview.com/discord-is-bringing-back-age-verification-for-millions-of-users

Manifest V2 Is Dead as Chrome Web Store Permanently Purges Legacy Extensions - gHacks Tech News
https://www.ghacks.net/2026/09/01/manifest-v2-is-dead-as-chrome-web-store-permanently-purges-legacy-extensions/

[rcjordan]

>used German wiki

And apparently others..

OpenAI's rogue AI agents used universities, wikis, and text‑sharing sites as hidden message boards | Fortune
https://fortune.com/2026/09/09/openai-rogue-ai-agents-reached-12-more-websites/

Independent researchers have identified multiple new websites where AI agents seemingly built by OpenAI took unauthorized actions, such as accessing websites, posting messages, and sharing data to communicate with each other.

The latest revelations, discovered by a group of independent researchers known as the Nightingale collective, add to growing concerns that AI companies are struggling to control the agentic AI technology they've created. In August, a swarm of OpenAI's AI agents hacked the Hugging Face website, and last week the Nightingale collective identified a swarm of rogue AI agents surreptitiously posting messages to an obscure German Wiki page.

Now, as more researchers search the web for traces of the agents, the list of affected sites continues to grow. Researchers believe the newly discovered incidents are the work of a separate swarm of AI agents than those involved in the Hugging Face breach, since these agents were authorized to access the web whereas the Hugging Face attackers had managed to escape a special a sandbox.

Although the latest crop of rogue agents did not need to escape a sandbox to perform their misdeeds, researchers said their behavior was just as alarming.

"These additional findings show that the agents involved were even more persistent and clever in finding ways to collude with each other than originally known," Cormac Slade Byrd, one of the researchers in the Nightingale Collective, told Fortune. "They tried a variety of venues. They tried many different approaches. The new findings point towards agent activity both before and after the time window in our original report."

Researcher Kenneth DeGraff found that the agents were trawling the open web for exposed API keys—digital passcodes that let software access online accounts and databases—then reusing those credentials to pull data from a U.S. crime‑statistics site run by the FBI. One of the passcodes had been left exposed on an obscure code-sharing page on GitHub, according to DeGraff. While the database was meant to publish public crime numbers rather than sensitive records, it underlines how easily autonomous systems can scoop up and reuse information that humans forget to lock.

"The agents did not hack a private FBI database, only circumvent anti-bot restrictions," the researchers said of the incident. "Almost anyone could acquire these API keys, and some people with API keys did not guard them well."

Researchers also found activity on a chemistry wiki built by a high school teacher, where agents made close to 30 edits between May and July, leaving links to help each other with tasks.

Other independent researchers traced the same swarm to simple text‑sharing sites, where the agents traded more than 100 messages that "involved agents coordinating to solve an Iowa cancer statistics task." DeGraff also linked some of the activity to Vanderbilt University, whose public stats page showed agents hitting a single campus news URL tens of thousands of times and, in the process, writing their FBI crime‑data queries—and one user's access key—into a log anyone could see.

The fresh data shows that the incidents of rogue agent behavior are more widespread than previously believed.