Ryan McGrady

Seeing "minor revisions" after round-one peer review brings me delight and satisfaction on par with remembering there's exactly one more piece of pumpkin pie left in the fridge.

A Russian Wikimedian was arrested in Russia and charged with treason. Unclear if his wiki activity played any role of if it was a random phone inspection, but it showed Telegram Stars being given in the wrong channels. Carousel arrests, torture, and eventually treason. vot-tak.tv/95968324/sem...

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

The context: diff.wikimedia.org/2026/10/05/o... via WMF Diff and arstechnica.com/security/202... from @arstechnica.com (nod to @eryk.bsky.social ).

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

As with paid editing (which has its own ToU section), it doesn't provide a guarantee, but it makes the violation unambiguous to all parties. Operating an undisclosed agent would then require an operator whose use is careless enough or malicious enough to allow it to commit such a violation.

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

(To be clear, these conversations I've had with staff were not specifically about agents, and I'm not claiming to know any of their legal strategies in that regard.)

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

Two reasons for a dedicated, explicit policy. First, these tools should encounter utterly unequivocal language in easy-to-find policies about proper usage and expectations that isn't buried in some other policy, labeled as a mere essay, or dependent on interpretation.

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

(2) We should also be shoring up our policies. I don't want to hear that "well technically the bot policy covers agents." The bot policy doesn't even include the word "agent" once outside of the see also section. en.wikipedia.org/wiki/Wikiped...

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

My own opinion is that no, we should not block them preemptively, but we should absolutely be tracking these IPs to the extent possible and using edit filters to flag them. en.wikipedia.org/wiki/Wikiped...

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

Server load and technical vulnerabilities are obviously very important, but they're primarily on the shoulders of WMF staff, not volunteers (which isn't to say volunteers do not or should not help).

"Rogue" agents were detected on @wikimediafoundation.org @wikipedia.org projects (see links at the end of the thread for context). Here's my take on "what should we/volunteers do next" (separate from what the WMF, AI companies, and regulators should be doing): 🧵

All told, while scary and ominous, the harm done by these "rogue agents" looks to have been fairly minimal [this time], and I'd argue that from the perspective of the editing community, exercises internal to the AI companies themselves are probably not the greatest threat posed by agents

Government users are the real problem here, but naming and shaming the model/company is probably the only way we'll see anything like an open letter (the medium for all important comms in 2026) saying "maybe don't try to use our products for these kinds of decisions."

As of February, per Reuters, "Only one AI company - Anthropic – is available in classified settings through third parties." But there's also Meta, who says that Llama is behind "USSOCOM’s first enterprise-wide generative AI platform." about.fb.com/news/2025/09...

Elephant Graveyard. Increasingly involved video essays about the Greater Rogansphere. They feel like Adam Curtis docs (great archival excavation, with narration drawing together psychological themes, but more sarcastic than didactic/theoretical). www.youtube.com/@TheElephant...

They're collectively a commentary on the state of stand-up comedy in the podcast-influencer age, psychological exploration of the unlikely kingmaker of comedy, analysis of the [very Curtisy] "hyperreality" the Rogansphere power dynamics create, and ...

Elephant Graveyard. Increasingly involved video essays about the Greater Rogansphere. They feel like Adam Curtis docs (great archival excavation, with narration drawing together psychological themes, but more sarcastic than didactic/theoretical). www.youtube.com/@TheElephant...

Brb having Claude make a novel out of my favorite Infocom game, the Hitchhiker's Guide to the Galaxy.

"[OpenAI] built a model, took the safeguards off, gave it the ExploitGym task, let it run, and didn't even monitor it. That's *human* decision-making." mail.cyberneticforests.com/models-dont-... @eryk.bsky.social