In the race to modernize third-party risk management, organizations have discovered a seductive shortcut: let AI handle the questionnaires. Vendors use generative AI to draft responses; security teams use AI to analyze them. The process is faster, cheaper, and appears more scalable than the manual alternative.
According to Gartner, this is precisely the problem.
Gartner’s research warns that GenAI could make questionnaire driven TPCRM increasingly disconnected from actual risk indicators rather than solving the underlying limitations of the process.
By 2028, Gartner predicts 70% of organizations and their vendors will use GenAI on both sides of the questionnaire process, a milestone that will make the entire exercise faster, more polished, and increasingly disconnected from actual risk .
The core issue is what Gartner calls “output degradation.” When AI generated content is analyzed by AI systems, errors can compound rather than cancel out. Each layer of automated interpretation introduces subtle distortions, assumptions, and artifacts that accumulate over time. The result is a gradual loss of signal: organizations believe they’re becoming more data driven while actually making decisions based on increasingly noisy and unreliable outputs .
It’s a photocopy of a photocopy, passed through another machine that’s been trained to describe photocopies.
The Example: A Perfectly Polished Fiction
Consider a mid-sized SaaS vendor that receives a 300 question security assessment from a prospective enterprise customer. The vendor’s security team, stretched thin and facing a tight sales deadline, feeds the questionnaire into an AI tool designed to auto complete security questionnaires based on the company’s existing documentation.
The AI generates fluent, confident responses. It maps answers to SOC 2 controls, ISO 27001 domains, and NIST framework references . Every answer sounds authoritative. The vendor submits the completed questionnaire.
On the buyer’s side, the security team runs the responses through their own AI powered analysis tool, which scores the vendor, flags potential gaps, and generates a risk summary. The vendor passes. The deal closes.
Six months later, a minor misconfiguration in the vendor’s cloud environment exposes customer data. The vendor had implemented the controls it described, but those controls had drifted. The questionnaire captured a snapshot of intent from a specific day. The AI analysis never questioned whether that snapshot reflected current reality .
This isn’t a hypothetical failure of technology. It’s a failure of model design. Questionnaires have always been point-in-time, self reported assessments. AI doesn’t change that fundamental limitation, it just makes the process more efficient . As one industry observer put it, “If your TPRM system starts with questions, AI will only help you get the wrong answers faster” .
The Evidence Problem
The deeper issue is that questionnaires, even AI enhanced ones, collect claims rather than evidence. A vendor can assert that it enforces multi-factor authentication across all privileged accounts. But an assertion is not a log. A policy statement is not a configuration file. A certification is not proof that controls are operating effectively today .
Research from the 2026 Verizon Data Breach Investigations Report reinforces the gap between compliance and security. Third parties were involved in 48% of breaches Verizon analyzed, a 60% increase from the previous dataset. More troublingly, only 23% of third-party organizations had fully remediated missing or improperly secured multifactor authentication on cloud accounts. Nearly half of weak password and permission misconfiguration findings took almost eight months to resolve .
A vendor can hold SOC 2 and ISO 27001 certifications and still carry active, exploitable risk. Compliance frameworks validate that processes exist. They do not validate that those processes are working right now.
The Mitigation: Continuous Monitoring Over Periodic Attestation
The path forward isn’t abandoning AI, it’s redirecting it toward problems that are actually solvable.
Gartner’s recommendation is unambiguous: prioritize continuous monitoring over questionnaire driven assessment . By 2028, the firm predicts half of all TPCRM programs will focus on continuous monitoring, allowing CISOs to repurpose due diligence resources toward higher value mitigation activities .
Continuous monitoring works by observing vendor risk as it evolves rather than asking vendors to describe it. This means pulling external attack surface data, threat intelligence feeds, breach disclosures, credential leaks, and vulnerability disclosures then correlating those signals against the vendor’s actual business criticality and data access .
The most effective programs combine external monitoring with internal telemetry. External tools identify exposed assets, malware infections, and breached credentials. Internal systems reveal unusual API activity, vendor account behavior, and privilege drift. Neither view alone is complete; together, they provide a dynamic picture that no questionnaire can match .
Critically, continuous monitoring requires trigger based oversight rather than calendar based reviews. A vendor that suffers a breach, leaks credentials, or adds a material subcontractor should automatically re-enter scrutiny. A vendor that simply passes another year should not be assumed safe .
Where AI Actually Helps
Gartner doesn’t advocate abandoning AI in third-party risk management. The firm argues for applying it where it creates genuine security value: scaling monitoring across large vendor ecosystems, detecting control drift earlier, and surfacing patterns that human teams might miss .
Used this way, AI becomes an instrument of observation rather than attestation. It watches for changes in vendor posture instead of processing claims about posture. It correlates signals instead of scoring narratives. It helps security teams prioritize investigation rather than produce paperwork.
The distinction matters because the alternative is worse than inefficiency. An organization that automates a broken process doesn’t fix the process, it industrializes its failure. Faster onboarding doesn’t mean better risk management; it means you can be wrong at scale .
The organizations that emerge from this transition won’t be the ones with the most sophisticated questionnaire automation. They’ll be the ones that stopped asking vendors to describe their security and started watching it instead.
Reader question: If your organization stopped sending security questionnaires tomorrow, what evidence would you actually need to trust a critical vendor and do you have systems in place to collect it continuously, or would you be flying blind?