The biggest crypto theft on record took about 20 minutes, and the rehearsal took longer than the robbery. Elsewhere, the AI boom keeps sending bills: Google admits AI search earns less, Amazon will spend $220 billion, and DoorDash’s agent customers barely exist.
1. The Binance heist was rehearsed for 36 hours, then took 20 minutes
The attackers tested their code on small wallets before touching Binance. TRM Labs says preparation began at least 36 hours before the theft. The operator rehearsed on a third wallet, then installed the same code on a Binance hot wallet four hours later. Around 4pm ET on October 29, funds started draining from wallets on at least ten blockchains. TRM puts most of the theft at about 20 minutes.
Binance first blamed a compromised engineer’s laptop that reached a service container and submitted unauthorized withdrawals. It later said the investigation had found “additional points of breach.” It says it has seen no evidence that private keys leaked, and that it will cover all losses. TRM, Chainalysis and Elliptic point to North Korea’s TraderTraitor group, though Elliptic stops short of naming it. Neither Binance nor law enforcement has attributed the attack.
The recovery numbers are the part founders should read twice. Elliptic says about $2.8 billion has been frozen, almost all of it stablecoins, which issuers can freeze. The other $3.3 billion is mostly Bitcoin, Ether and XRP, which nobody can freeze. It can be stopped only if it reaches an exchange. My reading: the assets that are most decentralized are also the ones with no emergency brake. Binance’s real question is why one compromised endpoint could reach systems that move billions.
TechCrunch · Ars Technica · WIRED
2. Google says AI search earns less, and its answer is more ads
Ars Technica’s headline puts it plainly: Google admits AI search makes less money, but it is fixing that with more ads. Ars adds that Google’s negative cash flow ballooned to -$9.06 billion and the stock fell after Q3 earnings. The Register’s take is that the search revenue “Ferrari” hit an AI-powered speed bump. We only have headlines and excerpts for both, so the mechanics of the revenue gap are not in evidence.
Even so, the shape of the problem is clear. Google is paying to build AI search and is monetizing it less well than the product it is changing. Its fix is to put more ads in. That is a fair business response, but it tests how much ad load users will tolerate in a product sold on direct answers. Anyone building an ad-funded AI product should note that the incumbent with the best ad machine in history is still working this out.
3. GM’s $220 million settlement shows the data broker is no shield
GM will pay $220 million to 36 states over allegations that it sold driving behavior and location data to data brokers. Those brokers, including LexisNexis and Verisk according to the earlier New York Times reporting, then sold it on to insurers. The data came from Smart Driver, a free feature that scored driving habits and seat belt use. GM ended it in 2024.
The detail that matters for builders is the route. GM didn’t sell directly to insurers, yet it still ended up paying, so selling to an intermediary doesn’t insulate you from what happens downstream. The settlement requires express consent before GM collects or shares driving and location data, plus a third-party assessment of its privacy program. That assessment is new compared with the earlier FTC and California deals.
GM can still share anonymized data with certain partners. And its new OnStar AI assistant comes with a warning that customers who decline data sharing will likely lose some features. The consent fight isn’t over; it’s moving into product design.
4. DoorDash’s AI agent boom is currently ‘several hundred’ orders
DoorDash CEO Tony Xu told analysts that third-party agents such as Muse, Instinct and Fo have sent “maybe several hundred orders in total.” For scale, DoorDash’s 2025 report put its average at 3 million deliveries a day. The reported timeframe is vague, so the exact ratio is unclear, but the gap is enormous.
Xu’s framing was careful: “you never want to fight gravity,” but “you don’t want to be delusional.” Customers still order mostly through DoorDash’s own app. The company is also building its own agents, including a text-to-order agent, a CLI agent, Ask DoorDash and an MCP for corporate ordering. That hedge matters. If agents do become the front door, DoorDash wants to be one of them, not just their supplier.
The deeper issue is control. Xu floated advertising inside third-party agents, while Amazon blocks personal agents and Walmart welcomes them. Agent startups depend on platforms that haven’t decided whether to cooperate. Several hundred orders suggests the platforms don’t yet feel pressure to decide.
5. Amazon’s trillion-dollar AWS bet needs 40% growth to hold for years
Andy Jassy told investors Amazon still expects about $220 billion in 2026 cash capex, mostly for AWS, and “meaningfully more” in 2027. AWS grew 39% last quarter and has a $184 billion run rate. Jassy says it can grow north of 40% in 2027 and 2028 and become a trillion-dollar business in five to ten years. Amazon shares fell about 6% in after-hours trading.
My arithmetic: compounding $184 billion at 40% for five years lands near $1 trillion. So the target isn’t just about demand showing up. It needs the growth rate to hold at a scale no company has operated at. The Register’s headline says AWS is installing an AI accelerator every 14 seconds, but we only have its headline, not the underlying detail.
For founders, the practical read is that AWS is betting capacity will be the scarce thing. That may be good for availability, but it doesn’t guarantee better pricing. Microsoft is also posting record cloud growth while spending heavily, so this looks less like one company running away with the market than several giants financing the same bet.