AI agent security incidents<\/a> have already arrived at scale, with 54% of respondents reporting a confirmed breach or a near-miss caught before harm. Only 32% give every agent its own scoped identity, 69% have credential sharing somewhere in the fleet, and just 30% sandbox their highest-risk agents. The dominant security stack is OpenAI guardrails, Azure, and Google Cloud controls, used as primary coverage by 82% of enterprises, while dedicated agent-security specialists barely register in single digits.<\/p>\nWhat this means for your business<\/h2>\n
The incident numbers sort cleanly by where your organization sits on credential hygiene. Enterprises with any credential sharing in the fleet reported an incident or near-miss at 63.5%, against 40.9% for those with fully scoped, per-agent identities. That 23-point gap is the clearest thing in this survey, and it frames the question the CISO needs to answer honestly: not whether the organization runs AI agents, but whether each of those agents has its own governed identity or is sharing keys with a dozen others.<\/p>\n
The structural problem the data exposes is what you could call provider-comfort capture. Enterprises have defaulted to the guardrails their model vendors ship, rate that stack 4.2 out of 5, and simultaneously report that 54% have already been hit and only 35% believe their defenses are ahead of AI-enabled attackers. That contradiction is stable because provider-native controls are genuinely easy to adopt and genuinely inadequate for the identity and isolation gaps that determine blast radius, the scope of damage when one agent is compromised. Observation and enforcement are present in roughly half of enterprises; the containment layer that limits damage when those controls fail is present in only 30%. A security posture built on watch-and-prevent, with no box around the highest-risk agents, is one that propagates single failures widely.<\/p>\n
The buying behavior in the data confirms the stack is provisional even if satisfaction scores don’t. Among enterprises that have had a confirmed incident, 52.6% plan to replace or add tooling within 90 days, against 14% of those untouched. What’s conspicuously absent from those purchase plans is identity specifically: only 12% of respondents include an agent-identity product anywhere in their consideration set, and that share barely moves even among credential-sharing organizations that have already been hit. The control the incident data most directly implicates is the one most missing from the renewal conversation, which means the next contract cycle is the right moment to ask whether the provider bundle is a security posture or just a comfortable default.<\/p>\n
Concept deep-dive: Non-human identity<\/h2>\n
Non-human identity refers to the credentials and access rights assigned to software agents, bots, and automated processes rather than to people. Think of it as a badge-and-keycard system, except the badge-holder is a piece of code that can act on your systems around the clock. When agents share a single credential, compromising or misconfiguring one exposes everything that credential can reach. Scoped, per-agent identity limits that exposure and makes forensics possible after an incident.<\/p>\n
Based on reporting from The agent security gap: 54% of enterprises have already had an AI agent incident, and most still let agents share credentials<\/a>, originally published 2026-07-16 15:02:00.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"Share with your CISO A June 2026 VentureBeat Pulse survey of 107 enterprises (100-plus employees) finds that AI agent security incidents have already arrived at scale, with 54% of respondents reporting a confirmed breach or a near-miss caught before harm. Only 32% give every agent its own scoped identity, 69% have credential sharing somewhere in […]<\/p>\n","protected":false},"author":1,"featured_media":5637,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[238],"tmauthors":[],"class_list":["post-5636","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai-data","tag-ciso"],"_links":{"self":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts\/5636","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/comments?post=5636"}],"version-history":[{"count":0,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts\/5636\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/media\/5637"}],"wp:attachment":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/media?parent=5636"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/categories?post=5636"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/tags?post=5636"},{"taxonomy":"tmauthors","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/tmauthors?post=5636"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}