Can people be taught to catch a fake face? New research says yes, but only if we stop leaning so hard on the tools built to do it for us. The findings point to something uncomfortable: the way we train people to look matters more than the software we point at the problem.

EA learned a similar lesson about limits, though its lesson came from angry players rather than a lab. After the community revolted over microtransactions in College Football 27, the studio backed down. The pushback was real, and the reversal followed.

Then there's Zelda. The story of the princess's first English voice actress is quieter than either of those, but its reach runs deeper than most fans ever realized. Her work shaped where an entire franchise went.

Send us feedback at [email protected].

Apple has filed a lawsuit against OpenAI alleging the systematic theft of hardware trade secrets, naming OpenAI's chief hardware officer, Tang Tan, a 24-year Apple veteran, as a defendant. According to the filing, Tan directed job candidates still employed by Apple to bring physical parts from the company to their interviews for what the suit describes as "show and tell" sessions designed to extract confidential information. A separate defendant, former Apple employee Chang Liu, is alleged to have stolen an Apple laptop upon departing for OpenAI. Apple stated in a representative's comment to CNBC that significant evidence recently emerged suggesting that individuals at OpenAI wrongfully took secret information about unreleased technologies, processes, and products, with specifics in the filing reaching down to proprietary metal-finishing techniques.

The timing is striking given that the two companies remain active partners, with ChatGPT baked into Apple's iPhone software. The lawsuit lands less than a month after OpenAI was reported to have threatened Apple with its own legal action over the terms of that partnership, a move that, in retrospect, looks like an extraordinary miscalculation. Apple also has reason to be sensitive about OpenAI's broader hardware ambitions, particularly following OpenAI's acquisition of the Jony Ive-founded design firm io, an effort widely seen as aimed at building a device to compete with the iPhone and staffed in no small part with Apple recruits.

If the allegations hold up, the consequences for OpenAI could extend well beyond legal fees, potentially derailing the company's hardware ambitions for years at the precise moment it has assembled the software foundation to support them.

Read More →

Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.

Try Wispr Flow free

EA's College Football 27 made history this week as the first entry in the series to launch on PC, and by most accounts, the game itself is genuinely excellent. Some are calling it the best the franchise has ever been. The problem was not on the field. EA quietly locked progression features in Road to Glory and Dynasty, two offline single-player modes, behind paid microtransactions. Those XP-scaling tools had previously been free, and the decision to monetize them the way a mobile or free-to-play title would struck players as a serious breach of trust for a full-priced game. Steam reviews turned sharply negative, with the game sliding toward a "Mostly Negative" rating as users piled on.

After several days of sustained review bombing, EA blinked. The company issued a statement acknowledging it had "missed the mark" and announced it would remove the paid progression options entirely over the weekend. "Your feedback on Road to Glory and Dynasty is that we've missed the mark with the introduction of paid progression options," EA Sports wrote, adding a promise of "greater transparency and communication" around live service features going forward. The reversal has already begun nudging the Steam rating back in a more favorable direction.

The episode leaves one loose end EA has not addressed: players who already purchased College Football Points before the rollback have little to show for their money. Whether EA follows through on its promised transparency, or whether this becomes a recurring pattern with each new entry, will be worth watching when College Football 28 comes around.

Read More →

Artificial intelligence has become so effective at generating convincing human faces that the familiar tells, too many fingers, warped jewelry, and smeared backgrounds are rapidly disappearing. Researchers from the University of Aberdeen, working alongside Australia's National University, have found that a roughly one-hour structured training session can push a person's ability to spot AI-generated faces from around 40 percent accuracy to nearly 80 percent, with some participants approaching perfect scores. Rather than teaching people to hunt for glitches, the training focused on six perceptual qualities that modern image generators still struggle to replicate consistently: unusually perfect facial symmetry, highly proportional features, above-average attractiveness, generic-looking facial structures, limited emotional expression, and faces that are oddly difficult to remember after looking away.

The stakes behind this research are considerable. Deloitte estimates cited in the study project AI-enabled deepfake fraud losses in the United States could reach £40 billion next year, up from roughly £12 billion in 2023. A widely reported case in Hong Kong saw scammers use a deepfake video call to persuade an employee to transfer £25 million, and an AI-generated LinkedIn profile was found to have infiltrated U.S. policy circles. The researchers also noted that AI systems remain less reliable when generating older faces, younger faces, and those of people from underrepresented ethnic groups due to biases in training data, gaps that may still offer useful signals to a trained human eye.

The broader implication is that human perception can be sharpened through the same mechanism AI itself relies on, repeated exposure to examples and gradual pattern recognition, suggesting that trained human judgment, rather than automated detection alone, may prove one of the more durable defenses as generative models continue to improve.

Read More →

SpaceX’s playbook for recovering a rocket booster generally involves legs, a precisely controlled vertical landing, and either a concrete pad or a drone ship. China just managed to pull off something similar, but in a slightly different way, and on July 10, it tested the method as well.

China launched the Long March 10B from the Wenchang Spacecraft Launch Site in Hainan Province on July 10, 2026, at 12:15 PM local time. About six minutes after liftoff, the rocket’s first stage separated and began a controlled descent, which was caught by a cable net system aboard a specially designed recovery vessel called the Linghangzhe. No landing legs, no touchdowns, and a net on a ship. That is a world first. No other space program has ever captured an orbital-class booster this way, and that speaks volumes.

China Aerospace Science and Technology Corporation confirmed that the primary mission succeeded, with the satellite payload reaching its designated orbit. Reusability is the entire reason why SpaceX can currently launch at roughly twice the rate of all Chinese rockets combined. The recovery of the Chinese rocket does not immediately close the gap, but it is the first credible sign that China is on the same path.

Read More →

Meta has pulled its Muse Image tool just three days after launching it, following a wave of criticism over how the feature handled user consent. The tool allowed anyone to generate AI images by @-mentioning a public Instagram account in Meta AI, effectively using that account owner's photos as a visual reference without their knowledge or agreement. While an opt-out mechanism existed, the default setting enrolled every public Instagram user automatically, meaning millions of people were part of the system simply by having a public profile.

The backlash was swift and came from multiple directions. Talent agencies, including CAA, one of the largest in the entertainment industry, were among those pushing back against the feature. In a statement to Puck News reporter Dylan Byers, Meta acknowledged that the rollout went wrong, saying the feature "missed the mark" and confirming it is "no longer available." The company said its original intent was to provide a useful creative tool while giving people control over whether their public content could be referenced, but the default opt-in design undercut that stated goal from the start.

The episode is a pointed reminder that launching AI tools with opt-out rather than opt-in defaults carries real consequences, particularly when those tools involve generating images of real, identifiable people. Whether Meta revisits a version of the feature built around explicit user consent remains to be seen.

Read More →

The AI tool you tried last quarter waited for a prompt, hallucinated a number, then asked if you'd like a summary.

Viktor opened a PR at 2am, rebased it against main, ran your test suite, and posted a note in #eng: "Two flaky tests in payments service, both pre-existing. Recommended merging after fixing them." Then drafted the customer reply for the support ticket the bug created.

That's 619K autonomous actions per day across 20,000+ teams. Not chat replies. Real work shipped to GitHub, Stripe, Linear, Notion, and 3,000+ other tools, from inside Slack and Microsoft Teams.

You don't supervise him any more than you supervise a senior engineer.

SOC 2 certified. Your data never trains models.

"It's what you probably originally thought AI was going to be when you first heard of it in sci-fi movies." Tyler, CEO.

Get Started for Free

Nvidia RTX 5090 SE may slot between the 5090 and 5080

Nvidia is giving away free GPU trading cards

Little House on the Prairie season 2 adapts the Plum Creek book

A PNG image can trick AI coding assistants into leaking data

Terrorist groups are using AI tools to make bombs

Valve is accused of inconsistent Steam censorship