DataGrail: 63.6% of AI Vendors Hide Subprocessors

Every enterprise software vendor now claims to have "AI built in." Far fewer will tell you which outside companies that AI actually talks to. DataGrail's fifth annual Privacy and AI Trends Report, published in May 2026, put that gap in hard numbers: of 2,400 business software providers advertising AI capabilities, 63.6% did not disclose the third party AI subprocessors handling customer data in their legal documentation. Pair that with a parallel audit of 5,000 popular websites, and the report paints a picture of an industry that has spent a decade building consent infrastructure for cookies while quietly reopening the same problem through AI features and tracking pixels.

Key Takeaways

• DataGrail's fifth annual Privacy and AI Trends Report audited 2,400 enterprise AI systems and 5,000 popular websites in a benchmark published in May 2026.
• 63.6% of AI vendors did not disclose the third party AI subprocessors touching customer data in their legal documentation, and 32.8% of AI systems engage in at least one high risk activity like automated decision making.
• Organizations with high levels of "shadow AI" reported average breach costs of $4.63 million, about $670,000 more than organizations with low or no shadow AI exposure.
• 63% of the 5,000 audited websites still ignore Global Privacy Control and other universal opt out signals, meaning most sites simply disregard a browser level do not track request.
• A June 2026 DataGrail follow up flags tracking pixels and session replay software as a fast growing source of civil litigation, with thousands of class actions filed and a California court ruling that plaintiffs may no longer need to prove a hack to sue.

What Did DataGrail's 2026 Report Find?

DataGrail's report found that the AI boom has outpaced the paperwork meant to govern it. The company scanned 2,400 leading enterprise software systems that advertise AI features and found that nearly two thirds failed to name the subprocessors, meaning the outside AI models, cloud providers, or data partners, that actually process the data flowing through those tools. That is not a hypothetical risk. DataGrail found that 32.8% of AI systems participate in at least one high risk activity, such as processing sensitive personal data or making automated decisions about customers, without corresponding documentation to show the risk is managed. The report also links the two problems financially: organizations with high levels of "shadow AI," meaning AI tools adopted without formal security or privacy review, reported average breach costs of $4.63 million, roughly $670,000 higher than organizations with minimal shadow AI use.

DataGrail CEO Daniel Barber summarized the year bluntly: "If there's one word that sums up data privacy in 2026, it's 'more': more regulation, more risk, more pressure. The only thing there isn't more of is privacy professionals to handle it." The report also counted 145 new AI related laws enacted by state legislatures in 2025 alone, meaning compliance teams are being asked to govern a moving technical target against a moving legal one. Read the full report from DataGrail for the complete methodology.

Why Are Websites Still Ignoring Opt Out Signals?

Because there is currently little cost to ignoring them. DataGrail's audit of 5,000 popular websites found that 63% fail to honor Global Privacy Control (GPC) and other universal opt out mechanisms, the browser level signal that is supposed to function like a legally binding "do not sell or share my data" request under California and other state privacy laws. That is the same core compliance failure DataGrail has documented in prior years' reports, and it lines up with the enforcement numbers: California recorded $4.3 million in public CCPA consent settlements in 2025, much of it tied to sites failing to honor exactly these signals. Meanwhile, data subject deletion requests have risen 567% since 2021, with 87% of all requests now being deletion requests, pushing DataGrail to estimate that a mid sized company with 5 million site visitors could spend $1.5 million a year handling these requests manually.

Why Are Tracking Pixels Suddenly a Legal Liability?

Because courts are starting to treat pixel based tracking as a wiretap, not just a privacy nuisance. DataGrail's June 2026 follow up report describes a California ruling that expanded the California Invasion of Privacy Act's (CIPA) private right of action beyond traditional data breaches, to cover situations where consumer data reaches an unintended recipient even without any hacking or intrusion. In the report's words, "civil plaintiffs may no longer need to demonstrate a hack or intrusion." Tracking pixels and session replay tools like Hotjar and FullStory silently capture what a visitor does on a page, then send that data to a third party, which plaintiffs' attorneys now argue is functionally identical to tapping a phone line. This litigation wave has already produced real payouts: DataGrail flags health, financial, and legal services pages as facing the highest exposure, since the sensitivity of the underlying data amplifies the legal claim, and courts remain split on standing even as more cases reach discovery. Gblock has covered this trend in more depth in its breakdown of the 2026 CIPA lawsuit wave.

What Does This Mean for Email Tracking?

The same mechanism DataGrail describes on websites, an invisible tracker silently reporting a visitor's activity to a third party without meaningful consent, is exactly how an email tracking pixel works in your inbox. A one pixel image embedded in a marketing or sales email loads the moment you open it, quietly telling the sender when you read the message, your approximate location, and what device and email client you used. It is a smaller scale version of the same consent failure DataGrail found across 5,000 websites, just delivered through Gmail instead of a browser. Extensions like Gblock block these pixel requests before they ever reach the tracking server, addressing the same category of silent, unconsented data collection at the one place most people check dozens of times a day: their inbox. For a deeper look at how this data collection holds up at scale, see this peer reviewed study of 44,000 emails showing how easily pixel blockers get evaded, and this Rutgers hospital breach study linking pixel use to higher breach rates.

What Should Compliance Teams Do Now?

Compliance and privacy teams should treat this report as a checklist, not just a benchmark.

1. Audit AI vendor contracts specifically for subprocessor disclosure language; if a vendor cannot name who processes your data downstream, that is now a documented industry wide gap.
2. Test your own website against Global Privacy Control signals directly, since a cookie banner is not the same as honoring an opt out signal, and regulators are treating them differently.
3. Inventory every tracking pixel and session replay tool running on pages that touch health, financial, or legal information, given that is where DataGrail and the courts agree exposure is highest.

The Bottom Line

DataGrail's 2026 benchmark makes it hard to argue that privacy compliance is a solved problem for either AI vendors or website operators. Nearly two out of three AI vendors are not fully disclosing who touches your data, roughly two out of three websites ignore a legally significant opt out signal, and tracking pixels that were once dismissed as a marketing nuisance are now the subject of multimillion dollar settlements. If you want to see what unconsented tracking looks like from the receiving end rather than the compliance report, check your own inbox. Every marketing email with an invisible pixel is running a smaller version of the exact experiment DataGrail just audited at scale.

Installing a tracker blocker like Gblock will not fix subprocessor disclosure or GPC compliance, but it will stop your own inbox from being part of the problem.