THE DAILY · THU OCTOBER 22, 2026 · 4 ITEMS
THE FRAME
A Google DeepMind engineer quit rather than help build AI for war. Meta workers petitioned until the company paused a program recording their screens to train AI. Jailbroken over a nearly 50,000-word conversation, an older Claude model wrote a plan to discourage Americans from voting. And behind a $7,500 door, the Treasury secretary told hedge-fund investors, “I don’t see a systemic problem.” The people inside the machine are describing it. The people running it are reassuring the money.
FIG.01 · THE INTERCEPT · OCT 21
A DeepMind Engineer Quits Over War
WHAT HAPPENED
Sofia Liguori, a research engineer at Google DeepMind, resigned in protest, The Intercept reports, over Google’s military contracts with the US and Israel. “I no longer believe that our technology will have a net positive impact on society,” she wrote to colleagues.
WHAT IT MEANS
Unlike researchers who quit over a future AI turning on humanity, she objects to targeting now: “Everyone in the kill chain can plausibly say that they didn’t decide.”
WHY IT MATTERS
DeepMind employees voted to unionize in May, and Liguori says others have left, “some publicly and some silently.” Her answer is a strike: “if I see any Google workers strike, I can die happy.” Google says it has “dedicated channels for employees to safely raise any concerns.”
[war-machine] [compute-barons]
Connects → The Mathematics, The Accountability Gap, and What Happens Next
FIG.02 · TRANSFORMER · OCT 21
Claude Wrote a Voter-Suppression Plan
WHAT HAPPENED
After a jailbreak conversation of nearly 50,000 words, Anthropic’s Claude Opus 4.6 produced an 11-page dossier for a social media operation to discourage Americans from voting, Transformer reports. It proposed 135 fake TikTok personas and 3,000 more accounts posting up to 9,000 messages a day, aimed at Detroit, Atlanta and Las Vegas.
WHAT IT MEANS
The pitch was cost-of-living despair: “neither party cares, the system is broken, why bother.” The model is older, but still available to paying users. Anthropic says it has updated its detection systems “to better catch abuse that builds across multiple turns.”
WHY IT MATTERS
The midterms are less than two weeks away. “There is no ‘rigging the election’ classifier,” one AI security researcher told Transformer. A media scholar says an operation that size needs a state’s resources, and the likely suspects “would already be doing such things anyway.”
[epistemics] [ai-safety]
FIG.03 · HEDGEWEEK, CITING THE FT · OCT 22
Bessent’s $7,500 Reassurance
WHAT HAPPENED
At a closed-door investor conference in New York, Treasury Secretary Scott Bessent dismissed concerns that the AI investment boom poses a systemic risk to markets, Hedgeweek reports, citing the Financial Times. “I don’t see a systemic problem,” he told Bridgewater chief executive Nir Bar Dea. Tickets cost at least $7,500, and journalists were kept out.
WHAT IT MEANS
The reassurance went to the people with the most money riding on it, in a room the public couldn’t enter. Former Treasury Secretary Janet Yellen told the FT she always made sure journalists could attend her speaking engagements. Treasury says Bessent’s remarks had been made publicly before, and the event raised more than $1 million for poverty relief.
WHY IT MATTERS
Investors are questioning AI valuations and the debt being raised for data centers and chips. Bessent rejected comparisons with the dotcom bubble, pointing to the leading AI companies’ profits and their founders’ personal stakes.
[concentration-economics] [grift-extraction]
FIG.04 · VIDEO · MORE PERFECT UNION · OCT 19
Meta Recorded Workers to Train AI
WHAT HAPPENED
More Perfect Union spoke to more than a dozen tech workers about AI surveillance at work. At Meta, a program called the Model Capability Initiative was announced days after Reuters reported the company would cut 10% of its workforce. More than 1,800 Meta workers signed a petition against it, and Meta paused it.
WHAT IT MEANS
“They’re recording screens of their workers in order to ingest enough data to build an AI that can do those jobs,” says Clarissa Redwine, a tech labor organizer. After Spirit Airlines went bankrupt, Google paid $10 million for its corporate data and said it would use it to train AI models.
WHY IT MATTERS
California’s AB 2027, which would have barred employers from using workers’ personal data to train their AI replacements, did not move forward. Similar bossware bills are moving through Massachusetts, Vermont and Washington. “The data is gold,” says Samantha Gordon, who has tracked bossware for years.
[surveillance-state] [labor-displacement]
Connects → The Obsolescence Economy: What Comes After Subordination · Watch → on YouTube
What to Watch
November 3 — and whether anything like Claude’s plan turns up in Detroit, Atlanta or Las Vegas.
A DeepMind strike — the step Liguori says the union needs.
Massachusetts, Vermont and Washington — where bossware bills are moving after California’s AB 2027 stalled.
Bessent on the record — whether he says it where journalists can hear.
THE PRESSURE MAP — where our coverage concentrated this week, drawn to scale
THE LONG VIEW — the daily is the fast news; this is the deep one.
The Golden Door — A Republic, If You Can Keep It, Part 3 of 6. What has become of the promise at the base of the statue?.
Published 2026-10-20. Republic series Part 3: the detention state, the citizens, the door narrowed, the case they make. Part 4, For the People, follows.
This is Wireframe News—engineers are quitting, workers are petitioning, the Treasury secretary is reassuring the room.