Specificity has announced a formal partnership with the Association for Authenticated Secure Programmatic Advertising (A2SPA) to bring cryptographic execution-time controls to agentic AI systems operating inside the adtech stack - a move that arrives as the industry quietly wrestles with just how much autonomous decision-making it's willing to hand over to machines it can't fully audit.

The integration centers on enforcing verifiable constraints at the moment an AI agent actually executes a task, not before, not after. That distinction matters more than it might sound. Most current approaches to AI governance in advertising rely on policy layers - documented rules, contractual obligations, platform terms of service - that sit outside the execution environment entirely. They're aspirational. Cryptographic controls built into runtime, by contrast, are technical facts.

Editorial Analysis: Our analysis suggests The gap between what AI agents are permitted to do on paper and what they can be proven to have done in practice represents the single largest unaddressed liability exposure in the modern adtech stack.

Agentic AI in adtech isn't theoretical anymore. Systems are already bidding, pacing budgets, adjusting creative weights, and flagging brand-safety violations without a human touching the keyboard between decisions. That autonomy creates efficiency. It also creates attack surface.

The specific threat vector here is subtle but serious. An AI agent operating across a supply chain - pulling signals from a DSP, querying a data clean room, writing bid requests - is touching multiple trust boundaries in rapid succession. If one of those boundary crossings goes unverified, a compromised or misbehaving agent can propagate bad decisions (or worse, malicious ones) deep into a campaign before anyone notices. Traditional logging catches it after the fact. Cryptographic attestation at execution time stops it at the gate.

A2SPA's framework essentially asks each agent to prove, cryptographically, that it's operating within its sanctioned parameters before it's allowed to act. Specificity's technology sits at that checkpoint. "We've spent years building the infrastructure to answer a simple question: is this agent doing exactly what it's supposed to be doing, right now, and can we prove it?" said Jordan Blum, Chief Technology Officer at Specificity. "Partnering with A2SPA means those proofs become interoperable across the ecosystem, not just inside our own stack."

Interoperability is the operative word. proprietary attestation that only works within a single vendor's environment solves a narrow problem. What A2SPA brings is a standards body willing to define what "verified agent behavior" actually means across publisher, advertiser, agency, and platform boundaries - and willing to enforce it through membership requirements rather than just goodwill.

"The adtech industry has a long and expensive history of confusing documentation with enforcement. Cryptographic attestation at runtime is the first genuinely technical answer to a problem the industry has been papering over with PDFs for a decade. The resistance will come fast, and it will come from exactly the players whose margins depend on nobody being able to check their math." - Priya Nambiar, Head of Programmatic Infrastructure at Caldwell Media Group

Here's the tension nobody in this industry wants to say plainly: a lot of the AI governance frameworks currently being celebrated in adtech are marketing documents dressed up as technical standards. They describe intended behavior. They don't enforce it. And the companies most enthusiastic about publishing them are often the ones with the least incentive to let an independent party verify their claims.

That's not cynicism. It's the structural reality of a supply chain where dozens of intermediaries each control a slice of the data, each have their own commercial interests, and none of them are obligated to open their execution environments to outside scrutiny. The fraud problem in programmatic advertising - still a multi-billion dollar annual drain - persists partly because verification has always lagged behind the speed of automated decisioning.

Cryptographic execution controls don't fix fraud directly. But they establish a foundation of provability that fraud, by definition, can't survive. "The reason bad actors thrive in automated systems is opacity," said Mara Chen, Director of Standards Development at A2SPA. "Our framework assumes that if you can't prove your agent behaved correctly at execution time, you don't get to claim it did. That's a different standard than the industry has held itself to, and it's going to create friction with people who've benefited from the ambiguity."

That friction is real. Expect pushback from mid-tier intermediaries who've built margin on the gap between claimed and verifiable performance. Expect some of that pushback to be dressed up as concerns about latency, implementation complexity, or competitive sensitivity around proprietary models.

Specificity is betting that buy-side pressure will overcome that resistance faster than most people expect. Advertisers who've watched AI agents make expensive mistakes - mispaced campaigns, brand-safety failures, bid inflation that served nobody but the exchange - are increasingly willing to pay a premium for systems that can demonstrate, not just assert, that their agents behaved correctly.

The partnership is targeting initial rollout with select DSP and SSP partners in Q3, with broader ecosystem integration planned through A2SPA's membership certification process. What that certification process ultimately demands of participants - and how strictly it's enforced - will determine whether this becomes a genuine infrastructure shift or just another logo on a press release.

  • For DSPs and SSPs: Certification through A2SPA's membership process will likely become a commercial prerequisite - not a nice-to-have - as large agency holding groups begin writing verified agent behavior requirements directly into platform contracts. Those who delay integration risk being quietly deprioritized in preferred partner programs before any formal mandate arrives.
  • For Brands and Advertisers: Execution-time attestation creates, for the first time, a technically defensible audit trail when an AI agent does something it shouldn't. That changes the liability conversation with vendors substantially - "the algorithm decided" stops being an acceptable answer when you can prove what the algorithm actually did.
  • For Mid-Tier Intermediaries: This is the part of the supply chain with the most to lose. Margin built on the opacity between reported and actual agent behavior doesn't survive a regime where behavior is cryptographically provable. Expect consolidation pressure on intermediaries who can't pass certification without exposing performance claims that haven't held up to scrutiny.
  • For Publishers: Verification works in both directions. Publishers who can demonstrate that their own AI-driven yield optimization and floor-pricing agents operated within certified parameters gain a credible differentiator with direct-sold and programmatic buyers who are increasingly skeptical of supply path integrity.

The Q3 rollout with select DSP and SSP partners will function less as a product launch and more as a stress test - specifically of whether A2SPA's certification requirements have enough teeth to survive contact with commercial pressure from the platforms it most needs to certify. If the first cohort of certified partners includes even two or three significant SSPs, the network effect accelerates quickly; buyers consolidate spend toward provable supply paths, and holdouts face a market access problem rather than just a compliance one. The harder and more consequential question is whether A2SPA structures its certification as a living standard with real revocation mechanisms, or allows it to calcify into exactly the kind of aspirational checkbox the industry already has too many of. How that governance question gets answered in the next 18 months will determine whether cryptographic attestation becomes programmatic infrastructure or programmatic wallpaper.