TL;DR: Yesterday, the US’ top auto safety regulator launched a special investigation into a Tesla crash that killed a 76-year-old woman in her Texas home last Friday. The driver blames the car’s automated driving system—Tesla says it was driver error. The incident is just the latest in a long list of safety concerns about the company’s’s self-driving tech.

What happened: There’s no confirmation yet of which driving assistance feature was being used (if one was at all), but the blue Tesla Model 3 veered into the victim’s house “at a high rate of speed,” according to local police. Tesla’s top AI exec claimed on X that the driver overrode self-driving by flooring the accelerator, reaching a speed of 73 mph—apparently info gleaned from the company’s own review of the crash data.

Elon Musk added that the driver’s claim makes “no sense” because Full Self-Driving (its most advanced tier of automated driving features, which is locked behind a subscription) “drives slowly through neighborhood streets.” So far, none of the claims on either side have been independently verified.

FSD’s recent troubles: Tesla’s recent safety record sows some doubt about Musk’s implication that FSD is too cautious to cause a high-speed crash. A demo last year from a software safety advocacy group showed a Tesla Model Y with FSD enabled repeatedly blowing past a stopped school bus (and hitting child-sized mannequins). Last week, Swedish regulators even pushed the EU to reject the rollout of FSD software because it can be configured to ignore speed limits. And Tesla’s own data labelers—who review footage to train the self-driving software—have raised serious concerns about the cars’ safety, per a Reuters report from May.

Tesla’s own numbers: The EV company, for its part, has often said that FSD is 10 times safer than human drivers. Reuters’s recent analysis, though, found that the company’s safety stats were sometimes derived from misleading comparisons—like pitting Tesla vehicles against the average car in the US, which tends to be older and doesn’t include the newest safety features that automakers now implement to reduce crashes.

Bottom line: Tesla is no stranger to federal probes into its cars’ safety. Since 2016, the National Highway Traffic Safety Association has opened over three dozen special crash investigations involving the company’s automated driving systems. —WK

How to tell if a Prime Day deal is actually a deal

Gone are the days of stampeding through Walmart to snag a half-price TV on Black Friday. Today, one of the biggest sales events of the year kicks off—Amazon Prime Day (it runs through Friday).

Before you buy that Apple Watch you’ve been eyeing, it’s worth checking the price history of products to see whether that “65% off” deal is a genuine all-time low or if its price was quietly jacked up last week to make today’s discount look fatter.

Checking on Amazon: Amazon has a built-in feature through its Rufus chatbot—simply ask Rufus for price history on a product’s page to check its chart. But the chatbot “isn’t always reliable and often misses context,” per Wired’s testing. It’s also Amazon grading its own homework.

Checking independently: The free price-tracking tool CamelCamelCamel does the same job, and it’s always a good idea to cross-reference prices with third-party sources. Paste the Amazon link or a product’s ASIN (found in the Product Information section) into the site’s search bar to find the product’s lowest-ever price, its average price, and the price fluctuations. Alternatively, install its Camelizer browser extension or the Keepa extension to use the same features without switching tabs. —LC

A notorious crypto bot that’s spent years fleecing Ethereum traders just got fleeced right back—to the tune of $7.5 million. The bot (called Jaredfromsubway.eth) is known for its “sandwich attacks”: It spots someone about to buy a token and leaps in line ahead of them to buy first, which nudges the price up. It then instantly dumps at the higher price and pockets the difference. The Jared bot has been a thorn in the side of the crypto community for years—it even reportedly sandwiched a trade by Ethereum co-founder Vitalik Buterin once.

Then, someone set the perfect trap. Per crypto security firm Blockaid, the attacker spent weeks deploying fake tokens dressed up to look like easy money. The bot, doing exactly what it was programmed to do, took the bait, authorizing the attacker to withdraw some tokens from its wallet in exchange for the fake tokens. The attacker then used these permissions to drain about $7.5 million in crypto from the bot’s wallet (it’s all super technical, but you can learn how they did it here). For a lot of crypto traders, this trap couldn’t have been sprung on a better bot. —WK

Chaos Brewing Meter: /5