OpenAI just did something you almost never see in corporate legal disputes: it published private iMessages and emails to publicly embarrass a plaintiff. The target? Apple. The occasion? A trade secrets lawsuit that OpenAI is calling "careless, aggressive, and oddly personal."
Here's the backstory. Apple filed suit last month against OpenAI and two former employees — Chang Liu and Tang Tan — alleging they walked out the door with confidential information about unreleased Apple products and technologies. Liu was an iPhone engineer. Tan spent 25 years at Apple, including a long run overseeing iPhone and Apple Watch hardware, before becoming OpenAI's chief hardware officer. Apple wants a preliminary injunction to immediately block all three parties from using, sharing, or even accessing any alleged confidential information while the case plays out.
OpenAI's response was not a legal filing. It was a blog post. And that's a deliberate choice worth paying attention to.
Rather than letting its lawyers handle this quietly, OpenAI went straight to the public with a counter-narrative and receipts to back it up. The company's core argument is that Apple is misrepresenting what actually happened with Liu's post-departure access to Apple's internal systems. Apple frames it as Liu exploiting an authentication vulnerability to access cloud storage after he left and downloading sensitive files. OpenAI frames it very differently — arguing that Apple's own employees reached out to Liu after he departed and asked for his help locating information, and that residual access to company systems after an employee leaves is a known, systemic problem Apple hasn't fixed.
That's a meaningful distinction. If accurate, it reframes Liu less as someone who went out of his way to steal data and more as someone Apple's own team kept pulling back into their systems even after he tried to leave cleanly.
It's also worth noting what OpenAI is not saying. The company isn't disputing that Liu had access to Apple files or that Tan held a senior role with access to sensitive product information. It's disputing intent and context — harder things to prove, but also the things that tend to matter most in trade secrets cases.
The move to go public with private communications is a calculated risk. OpenAI gets to shape the narrative before Apple does, potentially influencing how the press and the public — and maybe eventually jurors — understand the case. But it also signals a combative posture that could complicate settlement talks down the road.
The bigger picture here is that this lawsuit is really about something Apple cares deeply about: its hardware future. Apple has been aggressively building out its AI infrastructure, and losing senior hardware talent to the company it just partnered with on Apple Intelligence is a genuine competitive concern. Whether that concern rises to the level of a legitimate legal claim is now for the courts to decide — but OpenAI is clearly not going to let Apple set the terms of the public debate while that process unfolds.
AI
US AI deal gives 50,000 Ukrainian kamikaze drones autonomous targeting
A $400 drone has reportedly taken down Russian military helicopters worth up to $19 million each. Now imagine 50,000 of those drones that can hunt targets on their own without a human holding the controls.
That's not a hypothetical. Since mid-July, Ukrainian forces have been taking delivery of Shrike FPV drones built by Ukrainian manufacturer SkyFall, upgraded with autonomous targeting hardware and software developed by US company Auterion. The two companies plan to deliver 50,000 of these upgraded drones in the coming months, each equipped with what Auterion calls a Skynode S strike kit.
The way it works is deceptively simple. A human operator flies the drone into the general area of a target — within roughly half a mile — designates what they want to hit, and then hands off control to the AI. From that point, the drone tracks and guides itself into the target autonomously. The operator can still abort the strike or redirect to a different target as long as radio contact holds. But if that signal drops — because of terrain, buildings, or enemy jamming — the drone keeps going anyway.
That last part is the piece that matters most. GPS jamming and radio frequency interference have become standard tools on the modern battlefield in Ukraine. Drones that depend on either one are increasingly vulnerable to disruption before they reach their targets. Auterion's system sidesteps both problems entirely by relying purely on the drone's onboard camera and visual processing to track and home in on moving targets. No GPS. No continuous radio link required.
The underlying hardware running all of this costs $18. That's not a typo. Auterion built its flight control and terminal guidance system around a Western-made Arm system-on-a-chip at that price point, which is how you make autonomous targeting practical at the scale of 50,000 units.
Until now, AI-guided drones in Ukraine have mostly been the expensive, long-range variety — like the $5,000 Hornet drones developed by Perennial Autonomy, a company backed by former Google CEO Eric Schmidt, which fly deep into Russian territory to hit infrastructure and energy targets. The Shrike is a different category of weapon entirely: cheap, short-range, and built for the brutal close-quarters attrition happening at the front lines, where FPV drones have become the dominant cause of casualties on both sides.
Bringing autonomous targeting to that class of weapon is a significant shift. It reduces the cognitive load on Ukrainian operators who currently have to manually guide each drone to impact under fire, and it makes the weapons harder to defeat through jamming alone. It also raises genuinely uncomfortable questions about where human oversight ends in lethal autonomous systems — questions that military ethicists have been raising for years and that deployments like this are now forcing into the real world.
Source: Ars Technica
That's not a hypothetical. Since mid-July, Ukrainian forces have been taking delivery of Shrike FPV drones built by Ukrainian manufacturer SkyFall, upgraded with autonomous targeting hardware and software developed by US company Auterion. The two companies plan to deliver 50,000 of these upgraded drones in the coming months, each equipped with what Auterion calls a Skynode S strike kit.
The way it works is deceptively simple. A human operator flies the drone into the general area of a target — within roughly half a mile — designates what they want to hit, and then hands off control to the AI. From that point, the drone tracks and guides itself into the target autonomously. The operator can still abort the strike or redirect to a different target as long as radio contact holds. But if that signal drops — because of terrain, buildings, or enemy jamming — the drone keeps going anyway.
That last part is the piece that matters most. GPS jamming and radio frequency interference have become standard tools on the modern battlefield in Ukraine. Drones that depend on either one are increasingly vulnerable to disruption before they reach their targets. Auterion's system sidesteps both problems entirely by relying purely on the drone's onboard camera and visual processing to track and home in on moving targets. No GPS. No continuous radio link required.
The underlying hardware running all of this costs $18. That's not a typo. Auterion built its flight control and terminal guidance system around a Western-made Arm system-on-a-chip at that price point, which is how you make autonomous targeting practical at the scale of 50,000 units.
Until now, AI-guided drones in Ukraine have mostly been the expensive, long-range variety — like the $5,000 Hornet drones developed by Perennial Autonomy, a company backed by former Google CEO Eric Schmidt, which fly deep into Russian territory to hit infrastructure and energy targets. The Shrike is a different category of weapon entirely: cheap, short-range, and built for the brutal close-quarters attrition happening at the front lines, where FPV drones have become the dominant cause of casualties on both sides.
Bringing autonomous targeting to that class of weapon is a significant shift. It reduces the cognitive load on Ukrainian operators who currently have to manually guide each drone to impact under fire, and it makes the weapons harder to defeat through jamming alone. It also raises genuinely uncomfortable questions about where human oversight ends in lethal autonomous systems — questions that military ethicists have been raising for years and that deployments like this are now forcing into the real world.
Enjoyed this?
Get stories like this delivered every Tuesday — free.