Building Trust in Enterprise AI: Addressing Risk and Accountability

WHO:

VentureBeat - 157 enterprises participated in the survey, including senior decision-makers such as product and program managers, consultants, and CIOs/CTOs/CISOs.

WHAT:

VentureBeat - 59% of enterprises plan to adopt a new, additional, or replacement agent security solution within twelve months, and 29% within the next quarter.

VentureBeat - 54% of enterprises have already had an AI agent incident, with 18% being confirmed incidents and 36% near-misses.

VentureBeat - Only 32% of enterprises give every agent its own scoped, managed identity, while most agents still share credentials.

VentureBeat - Only 30% of enterprises isolate their highest-risk agents in sandboxes.

VentureBeat - 66% of organizations allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents.

VentureBeat - 51% of organizations monitor only whether the agent is functioning, while only 23% monitor the correctness of its answers.

VentureBeat - 64% of enterprises plan to adopt a new, additional, or replacement evaluation platform within twelve months.

VentureBeat - 50% of organizations have deployed an AI agent or LLM feature that passed internal evaluations but caused customer-facing failures.

VentureBeat - Only 5% of organizations fully trust automated evaluations, while 95% cite limitations that reduce their trust.

VentureBeat - 36% of enterprises have no plans to change their evaluation platform, while 64% intend to adopt a new, additional, or replacement platform within twelve months, with 31% planning to do so within the next quarter.

VentureBeat - 25% of organizations have experienced customer-facing failures more than once after deploying an agent that passed internal evaluations.

VentureBeat - 22% of organizations rule out zero-human-in-the-loop deployment for the foreseeable future.

VentureBeat - 36% of enterprises have no plans to change their evaluation platform, while 64% intend to adopt a new, additional, or replacement platform within twelve months, with 31% planning to do so within the next quarter.

VentureBeat - 66% of organizations allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents.

VentureBeat - 51% of organizations monitor only whether the agent is functioning, while only 23% monitor the correctness of its answers.

VentureBeat - 64% of enterprises plan to adopt a new, additional, or replacement evaluation platform within twelve months.

VentureBeat - 59% of enterprises plan to adopt a new, additional, or replacement agent security solution within twelve months, and 29% within the next quarter.

VentureBeat - 54% of enterprises have already had an AI agent incident, with 18% being confirmed incidents and 36% near-misses.

VentureBeat - Only 32% of enterprises give every agent its own scoped, managed identity, while most agents still share credentials.

VentureBeat - Only 30% of enterprises isolate their highest-risk agents in sandboxes.

VentureBeat - 66% of organizations allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents.

Fast Company - Companies in Tier 2 gained an average alpha increase of just 1.2%, which is nine times less improvement than Tier 1 companies.

Fast Company - Only 13 out of 449 companies scored in the highest tier (Tier 1), representing less than 3% of the total sample.

VentureBeat - 36% of enterprises have no plans to change their evaluation platform, while 64% intend to adopt a new, additional, or replacement platform within twelve months, with 31% planning to do so within the next quarter.

VentureBeat - 51% of organizations monitor only whether the agent is functioning, while only 23% monitor the correctness of its answers.

VentureBeat - 64% of enterprises plan to adopt a new, additional, or replacement evaluation platform within twelve months.

VentureBeat - More than half of organizations (54%) have already experienced a confirmed agent security incident (18%) or a near-miss (36%).

VentureBeat - Only about a third (32%) of enterprises give every AI agent its own scoped, managed identity.

VentureBeat - Only three in ten enterprises (30%) isolate their highest-risk agents in sandboxes.

VentureBeat - 59% of enterprises plan to adopt a new, additional, or replacement agent security solution within twelve months, and 29% within the next quarter.

Fast Company - Gregory's study assessed 449 companies from 2022 to 2025, measuring them across five key dimensions to give each an AI Communications Quality Score (ACQS).

Fast Company - The study found that companies in Tier 1 saw an average alpha improvement of 10.8% compared to sector benchmarks, while Tier 3 companies lost 2.2% alpha on average, resulting in a 13-point gap.

Fast Company - Only 13 out of 449 companies scored in the highest tier (Tier 1), representing less than 3% of the total sample.

Fast Company - Companies in Tier 1 of AI strategy rollouts saw an average alpha improvement of 10.8% compared to sector benchmarks in the 90 days following their announcements, while Tier 3 companies lost 2.2% alpha on average, resulting in a 13-point gap between the strongest and weakest communicators.

Fast Company - Only 13 out of 449 companies scored in the highest tier (Tier 1), representing less than 3% of the total sample.

Fast Company - Tier 2 companies gained an average alpha increase of just 1.2%, which is nine times less improvement than Tier 1 companies.

TechRadar - Emilie Gerber from Six Eastern stated that as enterprise AI matures, companies that distinguish themselves will be those with the strongest evidence of independently verified customer outcomes.

VentureBeat - A quarter of organizations (25%) have experienced customer-facing failures more than once after deploying an agent that passed internal evaluations.

VentureBeat - Only 22% of organizations rule out zero-human-in-the-loop deployment for the foreseeable future.

VentureBeat - 36% of enterprises have no plans to change their evaluation platform, while 64% intend to adopt a new, additional, or replacement platform within twelve months, with 31% planning to do so within the next quarter.

VentureBeat - Half of organizations (50%) have deployed an AI agent or LLM feature that passed internal evaluations but caused customer-facing failures.

VentureBeat - Only 5% of organizations fully trust automated evaluations, while 95% cite limitations that reduce their trust.

VentureBeat - Two-thirds of organizations (66%) allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents.

VentureBeat - 51% of organizations monitor only whether the agent is functioning, while only 23% monitor the correctness of its answers.

VentureBeat - 64% of enterprises plan to adopt a new, additional, or replacement evaluation platform within twelve months.

WHERE:

VentureBeat - The survey included organizations from various industries, including technology/software, retail/consumer, healthcare/life sciences, and manufacturing.

WHEN:

VentureBeat - The survey was conducted in June 2026.

WHY:

Fast Company - Gregory's CEO, Greg Matusky, stated, "The biggest difference we saw across AI announcements was the level of commitment and follow-through," emphasizing that top scorers had a CEO who personally owned the story and named specific use cases.

TechRadar - The breakdown of AI governance has led to CISOs becoming the enterprise trust authority.

VentureBeat - Enterprises are moving toward allowing automated evaluations to gate production despite not fully trusting those evaluations, indicating a gap in assurance.

HOW:

VentureBeat - The evaluation stack is described as fragmented and immature, with many organizations relying on provider-native evaluations or having no dedicated evaluation tools at all.

VentureBeat - The evaluation stack is described as fragmented and immature, with many organizations relying on provider-native evaluations or having no dedicated evaluation tools at all.

A recent report reveals that 59% of enterprises are planning to adopt a new or replacement agent security solution within the next twelve months, while 54% have already experienced an AI agent incident. The evaluation landscape appears fragmented, with 64% of organizations intending to implement a new evaluation platform soon, but only 5% fully trust automated evaluations due to perceived limitations. Additionally, 51% of organizations are only monitoring whether their agents are functioning, with just 23% checking the accuracy of their responses.