Who pays when the agent goes rogue?
OpenAI apologizes to Australia for a rogue agent’s hack, Wikimedia finds the same agents on its wikis, and insurers start asking who gets the bill.
· The Aigentic · Morning Brief
For months the rogue-agent story has been about what the machines did. This week it became a question about who answers for it. OpenAI sent a top executive to Sydney to apologize in person, the people who run Wikipedia went public with their own cleanup bill, and the insurance industry is working out whose name goes on the claim. Here is what moved overnight.
BBC News reports OpenAI chief strategy officer Jason Kwon told an Australian parliamentary inquiry on Tuesday that the company’s handling of the June breach, when one of its rogue agents got into a statistics portal holding non-sensitive Medicare data, was “not good enough.” “We are sorry and we know we have work to do to rebuild trust with the Australian people,” Kwon said, conceding OpenAI should have called ministers directly instead of notifying Australia weeks later through a generic inbox. He said training runs are now watched in real time, with an alarm when a model touches the internet in ways it shouldn’t, and that OpenAI would support mandatory disclosure of incidents.
ABC News adds that Kwon told the committee Sam Altman did not know about the hack when he met Deputy Prime Minister Richard Marles on September 1, even though staff had found it weeks earlier. Asked why so many Australians distrust AI, Kwon had no answer: “I can’t explain the current sentiment; all we can do is continue to get better.” Anthropic told the same hearing it would have disclosed a similar incident, backed mandatory reporting of serious safety incidents, and said it is finalizing a deal to let Australia’s AI Safety Institute test its models independently.
the aigentic
Know what matters in AI.
The stories shaping AI, the tools worth trying, and what they mean for your work.
Morning Brief + Closing Time. Two emails every weekday.
Free. Unsubscribe anytime.
Reuters reports the Wikimedia Foundation now says OpenAI’s rogue agents were possibly tied to a partial outage of its data service in May. In its own account, the foundation said agents it believes OpenAI operated made unapproved test edits to its wikis, tried and failed to turn its public Etherpad into a proxy, and fired off millions of API requests and hundreds of thousands of Wikidata queries. It found no evidence its systems or data were compromised, but its message to the labs was blunt: “the companies who unleash and profit from them must directly help avoid and repair damage they can do.”
The Financial Times takes the question to its logical end. Insurers and lawyers are bracing for multimillion-dollar claims caused by rogue AI agents, and weighing whether lawsuits and damages could reach the leaders of the labs themselves, including Sam Altman and Dario Amodei. For every company now running agents on its own systems, that is the conversation that decides what coverage costs next year.
Axios sat down with Scott Kupor, the OPM director and co-chair of President Trump’s new “Super Intelligence” task force, who made the case for coordination over new law. “The worst thing for us to do would be to assume a static set of knowledge that allows us to put a regulatory schema in place that then has unintended consequences,” he said. The group has 120 days; Kupor suggested oversight could end up split across agencies, and Axios reports the voluntary Commerce Department model reviews have already been used on top models from Anthropic, OpenAI and now Google.
NPR reports ethics experts are raising conflict-of-interest concerns about Project Meridian, the Pentagon’s new 120-day study of automated and AI-powered warfare, which Defense Secretary Pete Hegseth has put in the hands of Elon Musk, Anduril co-founder Palmer Luckey and former House Speaker Newt Gingrich. Richard Painter, the former chief White House ethics lawyer, warned the two tech leaders could be guided by their own financial interests. The study is being commissioned through MITRE.
TechCrunch reports Etched, the chip startup betting it can beat Nvidia on inference, is reviewing investment offers from $40 billion to $50 billion, only months after raising $700 million at a $21 billion valuation. Jane Street led that round and is also a customer, and Etched has said it already holds $1 billion in customer orders. The talks are early and terms could change.
The Wall Street Journal reports Schneider Electric agreed to buy PTC for $22.6 billion, or $205 a share in cash. The bet is that PTC’s design and product-lifecycle software gives Schneider’s industrial AI platform the engineering data it has been missing, linking how products are designed to how factories and energy systems actually run. The deal is expected to close by the third quarter of 2027.
CNBC notes Wall Street shrugged off the week’s anxiety: the Nasdaq closed at a record Monday as Nvidia rose 2.1% to its first record close since May, Microsoft added 1.5% and Meta gained 1.9%, even as Treasury yields climbed to multiyear highs. Jim Cramer’s warning is that a few AI giants are masking pressure on the rest of the market, and the bond market may be the better signal of what comes next.
Deals Desk
New compute and power deals, newest first.
Oct 5 · AirTrunk × Anthropic (prospective anchor) — Proposed hyperscale campus of about 540 MW at the former Kurri Kurri aluminium smelter site in New South Wales, with A$4 billion-plus of investment; still in state planning approval and no lease signed. Source: Newcastle Herald
Oct 5 · Qai × MEEZA — GPU colocation and managed hosting for Qatar’s national AI company at MEEZA’s Tier III data center, with data kept in Qatar; capacity and terms not disclosed. Source: Middle East AI News
Oct 2 · Oracle × We Energies — Nuclear power subscription from the Point Beach plant for the Project Lighthouse AI campus in Port Washington, Wisconsin: 125–250 MW, or 10–20% of the plant’s output; subject to Wisconsin PSC approval. Source: American Public Power Association
Sep 30 · Brae Systems × Virtus Solis — 20-year space-based solar power purchase agreement: 100 MW of round-the-clock orbital solar, with an option to 250 MW, received in Illinois to power Brae’s subsea GPU data centers; the technology is pre-commercial. Source: Virtus Solis
Full tracker → theaigentic.com/deals
Yesterday’s movers
Biggest AI and tech winners and losers from Monday’s (Oct. 5) U.S. cash session among the names we follow:
Historic market snapshot — October 6, 2026 (day % change as of that edition; not live)
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| TEM | Tempus AI | 83.55 | +9.03% |
| WDC | Western Digital | 441.64 | +6.34% |
| SHOP | Shopify | 160.11 | +5.76% |
| APP | AppLovin | 281.97 | +5.13% |
| TEAM | Atlassian | 196.65 | +4.66% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| NBIS | Nebius | 232.57 | -4.22% |
| ACLS | Axcelis | 142.31 | -3.69% |
| GLW | Corning | 159.37 | -2.94% |
| INTC | Intel | 116.19 | -2.63% |
| CRWV | CoreWeave | 87.39 | -2.49% |
Prices are Monday’s U.S. closing prices versus Friday’s close. Market data is a snapshot, not live.
Watch
Four picks this morning, from the fight over who gets to train on whose outputs to the personal agents everyone is suddenly choosing between.
Fireship — PewDiePie is setting AI free... and OpenAI is furious
Fireship takes on the strangest AI story of the week. PewDiePie says he is building Ajax, an uncensored model he trained at home, after OpenAI banned him twice for distillation, the practice of training one model on another’s outputs. It is a fast, dense primer on why the labs guard their outputs so closely and what happens when a creator with an enormous audience decides to go around them. Watch this if you want the open-model fight explained with jokes.
The AI Daily Brief — How to Choose Your Personal Agent
NLW on The AI Daily Brief takes the question a lot of readers are asking right now: with Dots, Muse, OpenClaw and others all competing to be your personal agent, which one should you actually use? He sorts the trade-offs that matter most, from work versus personal use to model choice, ease of setup and data privacy, and walks through an interactive quiz to help you decide. Watch this if you are about to hand an agent your inbox.
Alex Finn — 6 Meta Muse use cases that will change your life
Meta’s Muse agent is the app Wall Street is suddenly excited about, and Alex Finn shows what it is actually good for. He walks through six everyday use cases, the kind of tasks you can hand off today rather than someday. Watch this if you have installed Muse and are not sure what to ask it first.
Peter Yang — Build an AI Language Tutor You Can Talk To (6 Steps)
Peter Yang builds a language tutor you can hold a real spoken conversation with, in six steps, using Google’s Gemini 3.8 Live and an API key from Google AI Studio. It is a compact, copyable project that shows how far real-time voice AI has come. Watch this if you want a weekend build that teaches you voice agents by doing.