AI Trade Secret Employee

State of play.

• Apple v. OpenAI is the most consequential AI trade secret case now in federal court. Filed July 10, 2026 in the Northern District of California, the complaint alleges OpenAI orchestrated a systematic campaign to recruit Apple employees and extract confidential hardware designs, manufacturing processes, and supply chain strategies — with over 400 former Apple employees now at OpenAI and Tang Tan, formerly Apple's VP of Product Design, named as OpenAI's Chief Hardware Officer .
• The noncompete enforcement landscape has fractured at the federal level. The FTC formally withdrew its nationwide noncompete ban in February 2026, removing the regulation from the CFR; enforceability now depends entirely on a state-by-state patchwork, with five states voiding virtually all employee noncompetes and others setting income thresholds .
• Taiwan's courts are imposing criminal sentences under national security law for semiconductor trade secret theft. A former Tokyo Electron/TSMC employee received a 10-year sentence under the National Security Act, with Tokyo Electron's Taiwan subsidiary fined T$150 million and ordered to pay TSMC T$100 million in damages .
• Shadow AI adoption inside enterprises has created a structural trade secret exposure most companies cannot yet see. A 2025 Gartner survey found 69% of organizations suspect or have confirmed employees using prohibited generative AI tools, with one-third of employees admitting to sharing enterprise research or datasets through unsanctioned platforms .
• For counsel advising technology companies, AI developers, or any employer with a proprietary-information program, the practical baseline is that trade secret risk now runs on four simultaneous vectors: deliberate institutional solicitation of competitor IP (Apple v. OpenAI), departing-employee theft in the semiconductor sector, passive leakage through shadow AI tools, and AI training platforms soliciting employer-owned work product.

Where things stand.

• Apple v. OpenAI tests whether institutional solicitation of competitor IP during recruiting constitutes actionable misappropriation. The complaint alleges OpenAI encouraged job candidates to bring Apple prototypes and components to interviews, disclosed unreleased product details to third parties, and deceived Apple's manufacturing partners by falsely claiming Apple approval to access proprietary metal finishing technologies — framing this as a corporate-directed scheme, not individual employee misconduct .
• The injunctive relief Apple seeks — forced redesign of OpenAI's hardware — is the high-stakes remedy that will define the case's practical impact. Apple is seeking both destruction of confidential materials and a court order requiring OpenAI to rebuild its hardware without using Apple's proprietary information, a remedy that could delay OpenAI's hardware launch by years .
• The FTC's withdrawal of the noncompete ban shifts the entire enforcement burden to state law. With no federal floor, a single noncompete agreement may be enforceable in one state and void in another; the FTC has signaled continued case-by-case scrutiny in staffing and healthcare, but broad rulemaking is off the table under the current administration .
• Taiwan's National Security Act is the operative enforcement vehicle for semiconductor IP theft. The TSMC/Tokyo Electron prosecution was brought under "core national technologies" provisions — not a conventional trade secret statute — producing heavier criminal exposure and corporate liability even without demonstrated external leakage .
• Undisclosed executive financial ties to competitors are an active termination and liability trigger. The Jay Chen termination at Tokyo Electron — involving financial relationships with investment vehicles funding Chinese chip equipment rivals — illustrates that conflict-of-interest policies need to reach beyond employment relationships to investment and advisory arrangements .
• Shadow AI usage is pervasive at every organizational level, including the C-suite. The 2025 Gartner survey documented that 93% of executives report using unauthorized AI, with a majority of C-suite members unconcerned about the practice, and 68% of workers using ChatGPT at work deliberately concealing it from employers .
• AI training platforms are soliciting employees' prior work product, which may be employer-owned. Mercor's model of paying individuals for work contributed in prior roles creates a direct pipeline for inadvertent — or deliberate — trade secret exfiltration, raising unresolved questions under assignment-of-inventions clauses and work-made-for-hire doctrine .

Latest developments.

• Apple filed a federal trade secret complaint against OpenAI, Tang Tan, Chang Liu, and io Products, LLC in the Northern District of California, alleging a coordinated scheme to extract hardware IP through recruiting and interview processes .
• JD Supra analysis of the Apple complaint details allegations that OpenAI encouraged candidates to bring Apple prototypes to interviews, deceived Apple's manufacturing partners, and that over 400 former Apple employees now work at OpenAI — framing the case's high bar: proving OpenAI's institutional knowledge of and encouragement for the alleged conduct .
• FTC formally withdrew its nationwide noncompete ban in February 2026, removing the regulation from the CFR; Washington enacted a near-total ban effective June 30, 2027, and Tennessee established a $70,000 income threshold beginning July 1, 2026 .

Active questions and open splits.

• Whether institutional recruiting conduct can establish corporate liability for trade secret misappropriation. Apple v. OpenAI turns on whether Apple can prove OpenAI directed — not merely benefited from — the alleged extraction of proprietary information during the hiring process; the "institutional knowledge and encouragement" standard is a high bar with no settled precedent in the AI hardware context .
• What noncompete and confidentiality architecture is adequate when the federal floor has been removed. With the FTC ban withdrawn and state law fragmenting, employers operating across multiple states face a compliance matrix where a single agreement template is almost certainly inadequate — and the FTC's residual case-by-case enforcement posture in staffing and healthcare means legal risk persists even without a rule .
• Whether shadow AI usage constitutes a trade secret disclosure under existing confidentiality agreements. Most confidentiality and acceptable-use agreements were not drafted with generative AI platforms in mind; whether inputting proprietary data into ChatGPT or Claude triggers a breach — and whether the employer can demonstrate it — is unsettled .
• Who owns work product contributed to AI training platforms. Mercor's solicitation of former employees' prior work raises unresolved questions about whether assignment-of-inventions clauses, work-made-for-hire doctrine, or confidentiality obligations survive employment and attach to AI training contributions .
• Whether Taiwan's national-security-law framework will influence how other jurisdictions charge semiconductor IP theft. The TSMC/Tokyo Electron prosecution under the National Security Act rather than a trade secret statute produced heavier criminal exposure and corporate liability even without demonstrated external leakage — a template other jurisdictions may adopt or resist .
• What conflict-of-interest disclosure frameworks are adequate for executives with international exposure. The Jay Chen termination leaves open what ongoing disclosure obligations semiconductor equipment companies should impose on executives whose roles give them access to strategic or technical information, particularly where China-revenue dependence creates structural tension .

What to watch.

• Early motions practice in Apple v. OpenAI — whether Apple secures a preliminary injunction halting OpenAI's hardware development, and what the court requires Apple to show on the "institutional direction" element, will set the litigation posture for the next several years.
• Discovery in Apple v. OpenAI for evidence of internal OpenAI communications and hiring directives — any document production showing company-level instruction to solicit Apple IP would be the evidentiary linchpin of Apple's institutional-liability theory.
• Whether OpenAI's formal response to the Apple complaint — which had not been publicly filed at the time of the last reporting — reframes the case around individual employee conduct rather than corporate direction, and whether that framing gains traction at the pleading stage.
• State legislative activity on noncompetes — Washington's near-total ban takes effect June 30, 2027, and additional states are moving toward restriction, which will continue to compress the geographic scope of enforceable noncompete coverage for AI-sector employers.
• Litigation or regulatory action arising from Mercor's AI training solicitations — any employer suit against a former employee or Mercor itself for misappropriation of work product would be the first significant test of the ownership question.
• Appeals in the TSMC/Tokyo Electron case — whether the 10-year sentence and corporate damages award survive appellate review will determine the precedential weight of Taiwan's national-security-law approach to chip IP theft.