An AI system that solely solutions questions carries one sort of danger. An agent that may question a database, name an API, or push a change to manufacturing carries one other sort of danger altogether. Two bulletins on August 4 present traders and distributors racing to construct a safety class across the second variety: Anaconda acquired Enkrypt AI, and AI-agent safety startup Zenity raised $125 million.
Anaconda folds safety into the AI improvement stack
Anaconda introduced the acquisition of Enkrypt AI on August 4, 2026, including mannequin, agent and Mannequin Context Protocol (MCP) safety to a platform already used throughout the Python and data-science lifecycle. Monetary phrases weren’t disclosed. Enkrypt AI runs pre-deployment red-teaming throughout greater than 300 assault classes, real-time runtime guardrails in opposition to jailbreaks and knowledge leakage, and compliance automation for frameworks together with the NIST AI Danger Administration Framework and the EU AI Act. Anaconda stated Enkrypt AI’s merchandise, plans and help proceed unchanged for present clients, and described the corporate as vendor-neutral and cloud-agnostic. Enkrypt is an OpenAI compliance integration associate, and Anaconda stated an identical model-agnostic strategy for Anthropic is coming.
The deal extends a sample. Anaconda acquired workflow-orchestration firm Outerbounds in April 2026 and has individually added Kilo Code to its portfolio, shifting into agentic developer tooling. Packages and environments, orchestration, AI improvement, and agent and mannequin safety now sit inside a single vendor relationship reasonably than 4 separate buying selections. Anaconda says 95% of the Fortune 500 and greater than 52 million customers depend on its platform, a company-reported determine value noting as such reasonably than as independently audited market share.
MCP turns knowledge entry right into a safety boundary
Mannequin Context Protocol servers let AI purposes and brokers connect with instruments and exterior knowledge sources. The connection is the purpose of MCP, and it’s also the publicity. Every new connection can widen an agent’s permissions, its attain into proprietary knowledge, and the floor an attacker can goal.
Enkrypt scanned greater than 268,000 instruments throughout 25,000 MCP servers within the two months earlier than the acquisition and reported vulnerabilities affecting 73% of them, greater than 143,000 in complete. Anaconda repeated the figures in its acquisition announcement. Deal with them as vendor analysis from the corporate that now stands to promote the repair, not as an independently audited census of MCP deployments usually. Even with the caveat famous, the size of the discovering explains why consumers are asking a special query about AI brokers than they requested about earlier generations of software program: not simply what a mannequin says, however what it’s allowed to do subsequent.
Zenity’s $125 million says capital is following the management layer
SecurityWeek reported August 4 that Zenity raised $125 million in a Collection C spherical led by Norwest, bringing its complete funding to $180 million. Qumra Capital, SoftBank Imaginative and prescient Fund 2, Hitachi Ventures, LG Know-how Ventures and present traders Vertex Ventures, Third Level Ventures, DTCP and Intel Capital additionally participated. Zenity, based in 2021, employs greater than 230 individuals and says it has tripled income in every of the previous two years. CEO Ben Kliger framed the elevate round scale: “Enterprise AI is skyrocketing,” he stated, with organizations “selling AI brokers at velocity and adoption charges by no means seen earlier than.”
Zenity’s platform is constructed round three capabilities it calls Floor, Implement and Defend: discovering brokers and their posture, imposing coverage, and defending in opposition to runtime misuse. The corporate argues safety has to judge an agent on the choice level, weighing context and intent instantly earlier than an motion executes, throughout ecosystems together with Microsoft, OpenAI, Claude, Google’s Gemini and customized agent frameworks. Zenity describes its work as AI-agent safety, governance and runtime enforcement, a narrower and extra correct label than “autonomous protection,” which neither its product pages nor its funding protection use. Zenity will not be a substitute for id and entry administration, endpoint detection or data-loss prevention. It provides a layer particular to what an agent is making an attempt to do and whether or not it needs to be allowed to do it.
Adoption is sort of common, however the knowledge behind it isn’t prepared
Dun & Bradstreet’s 2026 AI Momentum Survey of 10,000 companies discovered that 97% report energetic AI initiatives and 56% plan to extend AI funding over the subsequent 12 months. Solely 5% say their knowledge is sufficiently able to help the work, and 60% report a minimum of some measurable return, together with 24% describing broad or robust returns. Learn collectively, the numbers describe a market the place nearly everybody has deployed one thing and nearly nobody has completed the groundwork beneath it.
The hole above is the place the Anaconda and Zenity bulletins land. Brokers earn their worth by reaching enterprise knowledge and programs. The identical connection that offers an agent enterprise context is the connection that creates publicity, and weak knowledge readiness compounds the issue: an agent working in opposition to poorly ruled knowledge is more durable to audit and more durable to belief with write entry.
What consumers ought to ask earlier than they signal
Safety and platform groups evaluating the AI-agent safety class have particular, answerable inquiries to put to any vendor. Can the platform uncover each agent in use, together with ones constructed outdoors official channels? Does it map every agent’s id, credentials, knowledge entry and gear permissions? Can it examine and block a software name earlier than it executes, reasonably than solely logging it after the actual fact? Does it work throughout fashions from multiple supplier, or does it lock a purchaser right into a single vendor’s agent ecosystem? And when a vendor cites vulnerability or exploitation statistics, as Enkrypt did with its MCP scan, has anybody outdoors the corporate validated the methodology?
Procurement for the class now belongs to multiple desk. CISOs, CIOs, knowledge and AI leaders, and platform engineering groups every have a stake in agent stock, tool-call inspection, knowledge lineage and audit proof, and RFPs more and more must replicate the shared possession reasonably than routing via a single safety finances line.
The perimeter has moved to actions, not prompts
Anaconda’s acquisition and Zenity’s funding spherical will not be proof that enterprises have solved AI-agent danger. They’re proof that consumers, and the traders betting on them, have accepted agent safety as a definite finances line reasonably than a subset of generic AI governance. The class will maintain consolidating as extra platform distributors resolve it’s cheaper to accumulate the aptitude than construct it.
My take is that the distributors who matter received’t be outlined by the scale of their funding spherical. They are going to be outlined by whether or not a management can cease an unsafe motion earlier than it runs, on infrastructure the customer already trusts, with out relying on a single mannequin supplier to remain trustworthy.
An AI system that solely solutions questions carries one sort of danger. An agent that may question a database, name an API, or push a change to manufacturing carries one other sort of danger altogether. Two bulletins on August 4 present traders and distributors racing to construct a safety class across the second variety: Anaconda acquired Enkrypt AI, and AI-agent safety startup Zenity raised $125 million.
Anaconda folds safety into the AI improvement stack
Anaconda introduced the acquisition of Enkrypt AI on August 4, 2026, including mannequin, agent and Mannequin Context Protocol (MCP) safety to a platform already used throughout the Python and data-science lifecycle. Monetary phrases weren’t disclosed. Enkrypt AI runs pre-deployment red-teaming throughout greater than 300 assault classes, real-time runtime guardrails in opposition to jailbreaks and knowledge leakage, and compliance automation for frameworks together with the NIST AI Danger Administration Framework and the EU AI Act. Anaconda stated Enkrypt AI’s merchandise, plans and help proceed unchanged for present clients, and described the corporate as vendor-neutral and cloud-agnostic. Enkrypt is an OpenAI compliance integration associate, and Anaconda stated an identical model-agnostic strategy for Anthropic is coming.
The deal extends a sample. Anaconda acquired workflow-orchestration firm Outerbounds in April 2026 and has individually added Kilo Code to its portfolio, shifting into agentic developer tooling. Packages and environments, orchestration, AI improvement, and agent and mannequin safety now sit inside a single vendor relationship reasonably than 4 separate buying selections. Anaconda says 95% of the Fortune 500 and greater than 52 million customers depend on its platform, a company-reported determine value noting as such reasonably than as independently audited market share.
MCP turns knowledge entry right into a safety boundary
Mannequin Context Protocol servers let AI purposes and brokers connect with instruments and exterior knowledge sources. The connection is the purpose of MCP, and it’s also the publicity. Every new connection can widen an agent’s permissions, its attain into proprietary knowledge, and the floor an attacker can goal.
Enkrypt scanned greater than 268,000 instruments throughout 25,000 MCP servers within the two months earlier than the acquisition and reported vulnerabilities affecting 73% of them, greater than 143,000 in complete. Anaconda repeated the figures in its acquisition announcement. Deal with them as vendor analysis from the corporate that now stands to promote the repair, not as an independently audited census of MCP deployments usually. Even with the caveat famous, the size of the discovering explains why consumers are asking a special query about AI brokers than they requested about earlier generations of software program: not simply what a mannequin says, however what it’s allowed to do subsequent.
Zenity’s $125 million says capital is following the management layer
SecurityWeek reported August 4 that Zenity raised $125 million in a Collection C spherical led by Norwest, bringing its complete funding to $180 million. Qumra Capital, SoftBank Imaginative and prescient Fund 2, Hitachi Ventures, LG Know-how Ventures and present traders Vertex Ventures, Third Level Ventures, DTCP and Intel Capital additionally participated. Zenity, based in 2021, employs greater than 230 individuals and says it has tripled income in every of the previous two years. CEO Ben Kliger framed the elevate round scale: “Enterprise AI is skyrocketing,” he stated, with organizations “selling AI brokers at velocity and adoption charges by no means seen earlier than.”
Zenity’s platform is constructed round three capabilities it calls Floor, Implement and Defend: discovering brokers and their posture, imposing coverage, and defending in opposition to runtime misuse. The corporate argues safety has to judge an agent on the choice level, weighing context and intent instantly earlier than an motion executes, throughout ecosystems together with Microsoft, OpenAI, Claude, Google’s Gemini and customized agent frameworks. Zenity describes its work as AI-agent safety, governance and runtime enforcement, a narrower and extra correct label than “autonomous protection,” which neither its product pages nor its funding protection use. Zenity will not be a substitute for id and entry administration, endpoint detection or data-loss prevention. It provides a layer particular to what an agent is making an attempt to do and whether or not it needs to be allowed to do it.
Adoption is sort of common, however the knowledge behind it isn’t prepared
Dun & Bradstreet’s 2026 AI Momentum Survey of 10,000 companies discovered that 97% report energetic AI initiatives and 56% plan to extend AI funding over the subsequent 12 months. Solely 5% say their knowledge is sufficiently able to help the work, and 60% report a minimum of some measurable return, together with 24% describing broad or robust returns. Learn collectively, the numbers describe a market the place nearly everybody has deployed one thing and nearly nobody has completed the groundwork beneath it.
The hole above is the place the Anaconda and Zenity bulletins land. Brokers earn their worth by reaching enterprise knowledge and programs. The identical connection that offers an agent enterprise context is the connection that creates publicity, and weak knowledge readiness compounds the issue: an agent working in opposition to poorly ruled knowledge is more durable to audit and more durable to belief with write entry.
What consumers ought to ask earlier than they signal
Safety and platform groups evaluating the AI-agent safety class have particular, answerable inquiries to put to any vendor. Can the platform uncover each agent in use, together with ones constructed outdoors official channels? Does it map every agent’s id, credentials, knowledge entry and gear permissions? Can it examine and block a software name earlier than it executes, reasonably than solely logging it after the actual fact? Does it work throughout fashions from multiple supplier, or does it lock a purchaser right into a single vendor’s agent ecosystem? And when a vendor cites vulnerability or exploitation statistics, as Enkrypt did with its MCP scan, has anybody outdoors the corporate validated the methodology?
Procurement for the class now belongs to multiple desk. CISOs, CIOs, knowledge and AI leaders, and platform engineering groups every have a stake in agent stock, tool-call inspection, knowledge lineage and audit proof, and RFPs more and more must replicate the shared possession reasonably than routing via a single safety finances line.
The perimeter has moved to actions, not prompts
Anaconda’s acquisition and Zenity’s funding spherical will not be proof that enterprises have solved AI-agent danger. They’re proof that consumers, and the traders betting on them, have accepted agent safety as a definite finances line reasonably than a subset of generic AI governance. The class will maintain consolidating as extra platform distributors resolve it’s cheaper to accumulate the aptitude than construct it.
My take is that the distributors who matter received’t be outlined by the scale of their funding spherical. They are going to be outlined by whether or not a management can cease an unsafe motion earlier than it runs, on infrastructure the customer already trusts, with out relying on a single mannequin supplier to remain trustworthy.
