Apple Sues OpenAI for Trade Secret Theft, EU Bans Infinite Scroll: AI Enters the Deep Waters of Legal Oversight

AI Competition Enters the Deep Waters of Law and Regulation

The AI world has been unusually eventful lately — from trade secret lawsuits between tech giants, to the EU cracking down on addictive interaction design, to strategic product pullbacks. Taken together, these developments paint a clear picture: the AI industry has moved beyond a purely technical race and entered a complex new phase where legal boundaries, regulatory governance, and business model transformation are all deeply intertwined.

Apple Sues OpenAI: The Hidden War for AI Hardware Talent

Apple has formally sued OpenAI, alleging that the company poached employees to obtain hardware secrets. The complaint names specific project codenames, hardware components, and proprietary metalworking techniques, and claims that departing employees took unannounced prototypes and supplier materials with them.

It's worth noting that these allegations remain one-sided claims in a complaint and have not yet been adjudicated. But the significance of this lawsuit extends far beyond the individual case — it will directly test the legal boundaries between talent mobility and trade secret protection in AI hardware competition.

The core legal framework for trade secret litigation in the United States is the Defend Trade Secrets Act (DTSA), enacted in 2016, which unified federal civil litigation around trade secrets. Plaintiffs must prove three things: the information has independent economic value, reasonable measures were taken to maintain its secrecy, and the defendant acquired it through improper means. "Improper means" covers not only outright theft but also employees departing with their former employer's confidential information. Silicon Valley's most famous precedent is the 2018 Waymo (Google) v. Uber case, which ultimately settled with Uber paying roughly $245 million in equity — fundamentally reshaping talent mobility norms in the autonomous vehicle industry and providing an important reference point for this current suit.

As AI software companies like OpenAI accelerate their push into hardware, direct collisions with hardware giants like Apple are inevitable. When top talent moves between the two, determining what counts as "legitimate knowledge transfer" versus "trade secret theft" will become an unavoidable challenge for the entire industry. The outcome of this lawsuit could set the key legal tone for the AI hardware talent war.

The EU Steps In: Regulators Take Aim at "Addictive Design"

The EU has ordered Meta to disable autoplay and infinite scroll features or face substantial fines. This marks a significant shift in regulatory focus: from scrutinizing content itself to targeting the underlying interaction mechanics that drive continuous user engagement.

Autoplay and infinite scroll are not accidental product features — they are carefully engineered engagement mechanisms based on the behavioral psychology principle of "variable ratio reinforcement," operating on the same logic as slot machines. Infinite scroll was invented by Aza Raskin in 2006; he later publicly expressed regret, estimating the design consumes roughly 200,000 hours of human attention every day. The EU's legal basis for this action is the Digital Services Act (DSA), which requires very large platforms to assess and mitigate systemic risks to users' mental health, with fines of up to 6% of global annual revenue for violations. This is the first time the DSA has explicitly targeted the interaction design layer — a landmark moment.

Disabling these two widely used "addictive design" features would significantly affect user time-on-platform, ad inventory, and the feedback loops of recommendation systems.

What makes this even more noteworthy is its precedent-setting potential. Historically, regulation has focused on content and data dimensions — misinformation, privacy breaches, and so on. This time, the EU has set its sights on the interaction logic of the product itself. If this governance approach proves effective, it could become a reference model for other regions seeking to regulate addictive design, forcing platforms to rethink the business ethics behind "user attention."

OpenAI's Two Fronts: Biosecurity and Browser Retreat

OpenAI has two notable strategic moves worth examining.

Biosecurity Bounty Doubled

OpenAI has upgraded its existing Bio Bug Bounty program into a long-term private initiative, doubling the maximum reward to $50,000. The program targets researchers with experience in AI red-teaming, security, or biosecurity, with a focus on finding methods that can broadly bypass biosecurity safeguards.

The context behind this move is worth unpacking. AI biosecurity bounty programs are an extension of "red teaming" — a concept that originated in military adversarial exercises, matured in cybersecurity, and has since been adapted to AI safety evaluation. What makes biosecurity particularly unique is that large language models could lower the barrier to accessing synthetic biology knowledge, enabling non-expert actors to obtain technical guidance for designing pathogens. President Biden's 2023 Executive Order on AI explicitly required frontier AI systems to undergo "dual-use biotechnology" safety testing before deployment. OpenAI's decision to double the reward and shift to a private invitation-only format signals that the vulnerabilities being sought now extend beyond publicly disclosable security boundaries into more sensitive capability assessment territory — reflecting frontier AI labs' intense vigilance around "dangerous capabilities."

GPT Atlas Browser Discontinued

At the same time, OpenAI announced it is halting development of GPT Atlas, ending the exploration of a standalone AI browser in less than a year. The computer operation and long-task capabilities Atlas was exploring are being folded into a new "work-focused product" framework, with users needing to migrate workflows to a more unified entry point.

This pullback reflects a clear product-focus logic: rather than maintaining a standalone browser, integrating core capabilities like Computer Use into the main product line creates a more consistent agent execution experience.

Claude Reflect: AI Starts "Debriefing" Your Collaboration Habits

Anthropic has launched Claude Reflect Beta, helping users review how they use AI. It can analyze usage themes, frequency, and task types over the past 1, 3, 6, or 12 months.

The value of this feature lies in pushing beyond simple usage statistics into deep reflection on human-AI collaboration patterns. Rather than passively using AI, users can now examine their own collaboration habits and optimize their workflows accordingly. This also signals the next competitive dimension in AI products: competing not just on model capability, but on the "meta-capability" of helping users build efficient human-AI collaboration habits.

Embodied AI Breakthrough: Robot Completes Live Minimally Invasive Surgery

A humanoid surgical robot completed a remote minimally invasive cholecystectomy on a live pig. To be clear, the surgery was remotely controlled by a surgeon — this was a preclinical feasibility trial, not an autonomously performed AI procedure.

Understanding this milestone requires some technical context. Robot-assisted surgery is not a new concept — the da Vinci Surgical System has been FDA-approved since 1999 and has now completed over 10 million procedures. The core breakthrough in this experiment lies in combining "telepresence" with 5G low-latency transmission: the surgeon's control console and the robot's execution end are physically separated, with millisecond-level command transmission over high-bandwidth, low-latency networks. Laparoscopic cholecystectomy is a procedure with moderate technical difficulty and well-established clinical validation, making it a sensible choice for balancing feasibility with medical significance. It's worth noting that moving from preclinical animal trials to human trials still requires rigorous FDA or CE regulatory review, typically taking 3 to 7 years.

Nonetheless, this remains an important milestone in robotic-assisted surgery's precision and remote control capabilities. It validates the technical feasibility of remote minimally invasive surgery and opens up greater possibilities for extending high-quality medical care across geographic boundaries.

Copyright Litigation: The New York Times Alleges OpenAI Deleted Training Logs

In its ongoing copyright lawsuit against OpenAI, The New York Times has alleged that OpenAI concealed or deleted logs related to training data, and is asking the court to sanction OpenAI.

Again, these claims remain one-sided allegations in ongoing litigation and await final judicial determination. The broader context of this lawsuit matters: The New York Times v. OpenAI (filed in December 2023) is one of the largest and most consequential AI copyright cases to date. The central dispute is whether mass-scraping copyrighted content for model training constitutes "fair use." The evidence preservation question around training logs is especially critical: if the plaintiff can prove the defendant destroyed evidence after litigation commenced, the court may invoke the principle of "spoliation" to issue an adverse inference instruction to the jury — often more devastating in effect than losing on the merits. The legality and traceability of training data has always been the Sword of Damocles hanging over generative AI, and the outcome of cases like this will profoundly shape future compliance practices around large model training data.

AI in Practice: ChatGPT Work's Productivity Multiplier

Several enterprise case studies demonstrate the real-world value of ChatGPT Work.

In Stampley's case, ChatGPT Work integrated context from Jira, GitHub, and other tools, enabling a single-person team to handle the workload previously requiring 4 to 5 people. In RingCentral's case, one account manager expanded the number of clients they could track from 6 during the pilot phase to approximately 80.

Additionally, ChatGPT Computer Use has seen a noticeable speed improvement and gained a new real-time progress display — users no longer need to stay on the main interface to continuously monitor an agent's execution status. These cases collectively illustrate that AI's value is making a genuine leap from "conversational assistant" to "productivity multiplier."

Other Notable Developments

• Google Real-Time Multimodal Planning: Combining a camera, Nano Banana, and Google Maps, users can display real-world scenes, generate diagrams, and find nearby businesses all within a single voice conversation — available globally for free.
• Home Compute Experiment: Sunrun is deploying AI compute nodes in ordinary homes and paying participants, attempting to supplement large data centers with distributed home energy — though heat dissipation, noise, maintenance, and privacy questions still await real-world pilot testing.
• Telecom Goes AI-Native: Deutsche Telekom is working with OpenAI on an "AI-native" carrier transformation, aiming to rebuild core operational processes rather than simply adding a chat interface.
• AI's Climate Cost: Microsoft's report shows carbon emissions of 34 million tons, up 25% year-over-year, with the increase driven primarily by AI data center expansion. This figure reveals a structural contradiction between tech companies' 2030 carbon-neutral pledges and the rapid expansion of AI infrastructure — according to the IEA, global data center electricity consumption will double by 2026. Some companies are investing in nuclear energy or purchasing carbon credits to bridge the gap, but critics argue this shifts rather than truly solves the emissions problem.

Conclusion: Law and Regulation — The "Institutional Guardrails" on AI's Sprint

Looking across these developments, the AI industry is advancing on multiple fronts simultaneously, with each intertwined with the others. On the technical front, embodied intelligence and multimodal capabilities continue to break new ground. On the commercial front, product lines are being restructured around sharper focus. And on the legal and regulatory front — whether it's Apple and OpenAI's trade secret battle or the EU's forceful intervention against addictive design — "institutional guardrails" are being installed on AI's high-speed sprint.

It's increasingly clear that defining legal boundaries and establishing regulatory frameworks will be just as important as technological progress in shaping the future landscape of AI.

Key Takeaways