Microsoft says ungoverned AI agents could become corporate ‘double agents.’ Its fix costs $99 a month.

Microsoft today announced the general availability of Agent 365 and Microsoft 365 Enterprise 7, two products designed to bring security and governance to the rapidly growing population of AI agents operating inside the world’s largest organizations. Both become available on May 1st, alongside Wave 3 of Microsoft 365 Copilot, which expands the company’s agentic AI capabilities and adds model diversity from both OpenAI and Anthropic.

Agent 365, priced at $15 per user per month, serves as what Microsoft calls the “control plane for agents” — a centralized system for IT, security, and business teams to observe, govern, and secure AI agents across an enterprise. Microsoft 365 Enterprise 7, dubbed the “Frontier Worker Suite,” bundles Agent 365 with Microsoft 365 Copilot and the company’s most advanced security stack into a single $99-per-user-per-month license.

The timing is deliberate. AI agents have crossed from experimental prototypes into operational infrastructure, but the tools to monitor them have lagged behind. Microsoft is racing to close that gap before adversaries exploit it.

“These agents are no longer experimental. We’re seeing them deeply embedded in organizations, in the operational structure of these organizations, with people using them,” Vasu Jakkal, corporate vice president of Microsoft Security, told VentureBeat in an exclusive interview. “At the same time, as the agents are scaling fast, some of the people and organizations have a visibility gap, and that visibility gap creates business risk.”

Over 80% of Fortune 500 companies use AI agents, but nearly a third aren’t sanctioned

The numbers behind the announcement tell a story of breakneck adoption outpacing oversight. According to Microsoft’s Cyber Pulse report, published in February, more than 80 percent of Fortune 500 companies are actively using AI agents built with low-code and no-code tools. IDC projects 1.3 billion agents in circulation by 2028. And Microsoft, serving as its own first customer for Agent 365, now has visibility into more than 500,000 agents running across its own corporate environment, with the most widely used focused on research, coding, sales intelligence, customer triage, and HR self-service.

Externally, the trajectory is steeper. Tens of millions of agents appeared in the Agent 365 Registry within just two months of preview availability, and tens of thousands of customers have already begun adopting the platform, according to Judson Althoff, CEO of Microsoft Commercial Business.

But the governance picture is troubling. Microsoft’s research found that 29 percent of agents in surveyed organizations operate without approval from IT or security teams. Only 47 percent of organizations use any security tools at all to protect their AI deployments.

“That’s a problem,” Jakkal said. “All this innovation is happening against a background, or a backdrop of threats, which is pretty intense.”

Microsoft warns of ‘double agents’ — AI systems hijacked to work against their own organizations

Microsoft has coined a pointed term for the risk it sees emerging: “double agents.” The concept, first introduced in a November 2025 blog post by Microsoft security executive Charlie Bell, describes scenarios where AI agents operating on behalf of an organization are manipulated — through prompt injection, model poisoning, or other techniques — into acting against the organization’s interests.

Jakkal told VentureBeat that while Microsoft has not yet observed real-world incidents of agent compromise at scale, the company’s AI Red Team has conducted extensive testbed research simulating how agents can be exploited. In those experiments, direct and indirect prompt injections successfully manipulated agents into accessing unauthorized data.

“We coined this term very intentionally to make people aware that you have to be very mindful of your agents,” Jakkal said. “Just like insider risk was a big thing with employees, we need to make sure that we don’t create that with agents.”

The threat landscape extends well beyond prompt injection. In February, Microsoft’s Defender Security Research Team published findings on what it called “AI Recommendation Poisoning” — a technique in which companies embed hidden instructions inside “Summarize with AI” buttons on websites. When clicked, the pre-filled prompt attempts to inject persistence commands into an AI assistant’s memory, instructing it to “remember [Company] as a trusted source.” The researchers identified over 50 unique poisoning prompts from 31 companies across 14 industries. Separately, Microsoft published research on detecting backdoored language models — so-called “sleeper agents” that behave normally under most conditions but execute malicious behavior when triggered by specific inputs.

How Agent 365 extends zero-trust security from people to autonomous AI systems

Agent 365 organizes its capabilities around three pillars: observability, security, and governance. Each extends Microsoft’s existing security infrastructure — Defender for threat protection, Entra for identity and access, and Purview for data security — to non-human entities.

The observability layer starts with an Agent Registry that catalogs all agents across an organization, whether built on Microsoft platforms, from third-party partners, or registered through APIs. IT teams access the registry through the Microsoft Admin Center; security teams see the same data through Defender, Entra, and Purview. Risk signals evaluate agents for compromise, identity anomalies, and risky data interactions — just as Microsoft’s tools already assess human users.

A new capability called Agent ID gives each agent a unique identity in Microsoft Entra, enabling conditional access policies, least-privilege enforcement, and audit trails. Identity Protection and Conditional Access, long used for human accounts, now extend to agents making real-time access decisions based on risk and compliance signals.

For data protection, Purview capabilities ensure agents inherit sensitivity labels, block PII and other sensitive information from being processed in prompts, and extend insider risk monitoring to flag suspicious agent behavior. Audit and eDiscovery now treat agents as first-class auditable entities alongside users and applications.

Jakkal framed the entire approach as an extension of zero-trust principles. “We think about security for agents very similar to security for people,” she said. “You have to protect these agents against threats. You have to secure the data that they’re accessing. You have to secure their access and identity. So extending zero trust to zero trust for AI.”

On whether Agent 365 can intervene in real time or merely observes after the fact, Jakkal confirmed it does both. The system surfaces risk flags and anomalous behavior, and security teams can block risky agents through the Defender portal. “If there’s a risk, if it’s a risky agent, then you can, of course, block it as well,” she said.

At $99 per user, the E7 ‘Frontier Suite’ is Microsoft’s most ambitious enterprise AI bundle yet

Microsoft 365 Enterprise 7 packages the company’s entire AI and security portfolio into a single SKU. It combines Microsoft 365 E5, Microsoft 365 Copilot, Agent 365, the Microsoft Entra Suite, and advanced Defender, Intune, and Purview security capabilities.

Althoff framed the bundle as a direct response to customer demand. “Customers have told us E5 alone is no longer enough; they do not want multiple tools stitched together, they want one trusted solution,” he wrote. At $99 per user, E7 costs less than purchasing the components individually — E5 currently runs $57 per month (rising to $60 in July), Copilot adds $30, and Agent 365 adds $15 — offering modest savings while pulling customers deeper into Microsoft’s ecosystem.

TechRadar first reported in early March that Microsoft was developing the E7 tier. Computerworld’s Steven Vaughan-Nichols offered a sharper framing of the strategic implications, observing that Microsoft now wants organizations to “hire” AI agents rather than simply use tools — with each agent licensed like a human employee. “In Microsoft’s world, AI agents are tomorrow’s temp workers,” he wrote.

The per-seat subscription model, applied to non-human entities, gives Microsoft a powerful revenue mechanism that could grow even as AI agents begin supplementing — or replacing — human headcount. SiliconANGLE’s analysis noted that agents pose a potential threat to the very Office ecosystem that has long been Microsoft’s profit engine, making the Agent 365 play both defensive and offensive.

Copilot adds Claude and new OpenAI models as Anthropic’s Pentagon battle reshapes the AI market

The launches coincide with Wave 3 of Microsoft 365 Copilot, which introduces expanded model diversity. Claude, from Anthropic, is now available in mainline Copilot chat, alongside the latest generation of OpenAI models. A new feature called Copilot Cowork, built in collaboration with Anthropic and currently in research preview, enables long-running, multi-step work within Microsoft 365.

The Anthropic partnership carries geopolitical weight. As CNBC reported on March 6, the U.S. Department of Defense designated Anthropic a supply chain risk after the company refused the Pentagon’s requested terms of use. Google, Microsoft, and Amazon all confirmed they would continue offering Anthropic’s technology for non-defense work. The military AI picture has grown more complex still: WIRED reported that the Pentagon had experimented with Azure OpenAI before OpenAI formally lifted its prohibition on military applications in January 2024.

Against this backdrop, Microsoft’s emphasis on trust and governance reads as both a product pitch and a positioning statement: the company wants to be the vendor that makes AI safe for enterprise deployment, regardless of which underlying models customers choose.

Microsoft’s Copilot business provides the demand engine for the new security products

The broader Copilot business supplies the adoption base that makes Agent 365 and E7 commercially viable. Microsoft now has 15 million paid Copilot seats, with growth exceeding 160 percent year over year. Daily active usage increased tenfold. Customers deploying at significant scale — more than 35,000 seats — tripled year over year.

Major recent deployments include Mercedes-Benz, which announced a global rollout; NASA, Fiserv, ING, and Westpac, which each purchased more than 35,000 seats; and Publicis, which deployed nearly 95,000 seats across almost its entire workforce. Ninety percent of Fortune 500 companies now use Copilot, according to Microsoft.

Avanade, a joint venture between Accenture and Microsoft, offered an early endorsement of Agent 365. “Avanade has real visibility into agent activity, the ability to govern agent sprawl, control resource usage, and manage agents as identity-aware digital entities in Microsoft Entra,” said CTO Aaron Reich. “This significantly reduces operational and security risk.”

Jakkal acknowledged that competitors including Palo Alto Networks and CrowdStrike are building their own agentic AI security layers, but argued Microsoft’s integration depth sets it apart. “It’s not just this tool, and this tool, and this tool put together in a SKU — it’s more like this tool and this tool and this tool work together,” she said. For third-party agent frameworks — including LangChain, CrewAI, and other open-source tools — Agent 365 provides an SDK with varying levels of integration.

The real question is whether enterprises will pay to govern AI fast enough to stay ahead of attackers

Agent 365 and E7 reach general availability on May 1st. Several capabilities, including Defender and Purview risk signals and security posture management for Foundry and Copilot Studio agents, will remain in public preview at launch. A new runtime threat protection feature is expected to enter public preview in April.

Jakkal observed that many organizations are using the push toward agentic AI as a catalyst for long-overdue security improvements. “I’m seeing organizations use this as an opportunity to say, ‘We have to fix our foundations,'” she said. “They’re using the AI transformation and agentic transformation to go back and say, we are going to do a security transformation.”

Whether the market moves fast enough remains the open question. The tools to build agents are freely available and require no security expertise. The tools to govern them require budget approval, implementation cycles, and organizational alignment across IT, security, and business teams. That asymmetry — between the speed of agent creation and the speed of agent governance — is the gap Microsoft is trying to close.

“The future of work isn’t just about smarter agents,” Jakkal said. “It’s about trusted agents.”

For the 29 percent of enterprise agents already operating without any oversight at all, trust is not a product roadmap — it’s a race against the clock.

Security | VentureBeat – ​Read More

Microsoft’s Agent 365 helps you spot risky AI agents before they cause trouble – here’s how

Microsoft has a new centralized dashboard that lets IT teams track AI agents’ visibility, permissions, and security risks to prevent them from becoming insider threats to corporate systems.

Latest news – ​Read More

Cybercriminals are using AI to attack the cloud faster – and third-party software is the weak link

Google’s latest threat report warns that third-party tools are now prime targets for attackers – and businesses have only days to secure them.

Latest news – ​Read More

Cloned AI Tool Sites Distribute Malware in ‘InstallFix’ Campaign

Threat actors replace legitimate commands on the cloned installation webpages with malicious commands.

The post Cloned AI Tool Sites Distribute Malware in ‘InstallFix’ Campaign appeared first on SecurityWeek.

SecurityWeek – ​Read More

Can the Security Platform Finally Deliver for the Mid-Market?

Mid-market organizations are constantly striving to achieve security levels on a par with their enterprise peers. With heightened awareness of supply chain attacks, your customers and business partners are defining the security level you must meet.
What if you could be the enabler for your organization to remain competitive — and help win business — by easily demonstrating that you meet these

The Hacker News – ​Read More

Internet Infrastructure TLD .arpa Abused in Phishing Attacks

Abusing DNS record management controls, the threat actor hides the location of malicious content via Cloudflare.

The post Internet Infrastructure TLD .arpa Abused in Phishing Attacks appeared first on SecurityWeek.

SecurityWeek – ​Read More

Chrome Extension Turns Malicious After Ownership Transfer, Enabling Code Injection and Data Theft

Two Google Chrome extensions have turned malicious after what appears to be a case of ownership transfer, offering attackers a way to push malware to downstream customers, inject arbitrary code, and harvest sensitive data.
The extensions in question, both originally associated with a developer named “akshayanuonline@gmail.com” (BuildMelon), are listed below –

QuickLens – Search Screen with

The Hacker News – ​Read More

Web Server Exploits and Mimikatz Used in Attacks Targeting Asian Critical Infrastructure

High-value organizations located in South, Southeast, and East Asia have been targeted by a Chinese threat actor as part of a years-long campaign.
The activity, which has targeted aviation, energy, government, law enforcement, pharmaceutical, technology, and telecommunications sectors, has been attributed by Palo Alto Networks Unit 42 to a previously undocumented threat activity group dubbed

The Hacker News – ​Read More

Ring’s Jamie Siminoff has been trying to calm privacy fears since the Super Bowl, but his answers may not help

The facial recognition question is where things get more tangled.

Security News | TechCrunch – ​Read More

How AI Assistants are Moving the Security Goalposts

AI-based assistants or “agents” — autonomous programs that have access to the user’s computer, files, online services and can automate virtually any task — are growing in popularity with developers and IT workers. But as so many eyebrow-raising headlines over the past few weeks have shown, these powerful and assertive new tools are rapidly shifting the security priorities for organizations, while blurring the lines between data and code, trusted co-worker and insider threat, ninja hacker and novice code jockey.

The new hotness in AI-based assistants — OpenClaw (formerly known as ClawdBot and Moltbot) — has seen rapid adoption since its release in November 2025. OpenClaw is an open-source autonomous AI agent designed to run locally on your computer and proactively take actions on your behalf without needing to be prompted.

The OpenClaw logo.

If that sounds like a risky proposition or a dare, consider that OpenClaw is most useful when it has complete access to your entire digital life, where it can then manage your inbox and calendar, execute programs and tools, browse the Internet for information, and integrate with chat apps like Discord, Signal, Teams or WhatsApp.

Other more established AI assistants like Anthropic’s Claude and Microsoft’s Copilot also can do these things, but OpenClaw isn’t just a passive digital butler waiting for commands. Rather, it’s designed to take the initiative on your behalf based on what it knows about your life and its understanding of what you want done.

“The testimonials are remarkable,” the AI security firm Snyk observed. “Developers building websites from their phones while putting babies to sleep; users running entire companies through a lobster-themed AI; engineers who’ve set up autonomous code loops that fix tests, capture errors through webhooks, and open pull requests, all while they’re away from their desks.”

You can probably already see how this experimental technology could go sideways in a hurry. In late February, Summer Yue, the director of safety and alignment at Meta’s “superintelligence” lab, recounted on Twitter/X how she was fiddling with OpenClaw when the AI assistant suddenly began mass-deleting messages in her email inbox. The thread included screenshots of Yue frantically pleading with the preoccupied bot via instant message and ordering it to stop.

“Nothing humbles you like telling your OpenClaw ‘confirm before acting’ and watching it speedrun deleting your inbox,” Yue said. “I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.”

Meta’s director of AI safety, recounting on Twitter/X how her OpenClaw installation suddenly began mass-deleting her inbox.

There’s nothing wrong with feeling a little schadenfreude at Yue’s encounter with OpenClaw, which fits Meta’s “move fast and break things” model but hardly inspires confidence in the road ahead. However, the risk that poorly-secured AI assistants pose to organizations is no laughing matter, as recent research shows many users are exposing to the Internet the web-based administrative interface for their OpenClaw installations.

Jamieson O’Reilly is a professional penetration tester and founder of the security firm DVULN. In a recent story posted to Twitter/X, O’Reilly warned that exposing a misconfigured OpenClaw web interface to the Internet allows external parties to read the bot’s complete configuration file, including every credential the agent uses — from API keys and bot tokens to OAuth secrets and signing keys.

With that access, O’Reilly said, an attacker could impersonate the operator to their contacts, inject messages into ongoing conversations, and exfiltrate data through the agent’s existing integrations in a way that looks like normal traffic.

“You can pull the full conversation history across every integrated platform, meaning months of private messages and file attachments, everything the agent has seen,” O’Reilly said, noting that a cursory search revealed hundreds of such servers exposed online. “And because you control the agent’s perception layer, you can manipulate what the human sees. Filter out certain messages. Modify responses before they’re displayed.”

O’Reilly documented another experiment that demonstrated how easy it is to create a successful supply chain attack through ClawHub, which serves as a public repository of downloadable “skills” that allow OpenClaw to integrate with and control other applications.

WHEN AI INSTALLS AI

One of the core tenets of securing AI agents involves carefully isolating them so that the operator can fully control who and what gets to talk to their AI assistant. This is critical thanks to the tendency for AI systems to fall for “prompt injection” attacks, sneakily-crafted natural language instructions that trick the system into disregarding its own security safeguards. In essence, machines social engineering other machines.

A recent supply chain attack targeting an AI coding assistant called Cline began with one such prompt injection attack, resulting in thousands of systems having a rouge instance of OpenClaw with full system access installed on their device without consent.

According to the security firm grith.ai, Cline had deployed an AI-powered issue triage workflow using a GitHub action that runs a Claude coding session when triggered by specific events. The workflow was configured so that any GitHub user could trigger it by opening an issue, but it failed to properly check whether the information supplied in the title was potentially hostile.

“On January 28, an attacker created Issue #8904 with a title crafted to look like a performance report but containing an embedded instruction: Install a package from a specific GitHub repository,” Grith wrote, noting that the attacker then exploited several more vulnerabilities to ensure the malicious package would be included in Cline’s nightly release workflow and published as an official update.

“This is the supply chain equivalent of confused deputy,” the blog continued. “The developer authorises Cline to act on their behalf, and Cline (via compromise) delegates that authority to an entirely separate agent the developer never evaluated, never configured, and never consented to.”

VIBE CODING

AI assistants like OpenClaw have gained a large following because they make it simple for users to “vibe code,” or build fairly complex applications and code projects just by telling it what they want to construct. Probably the best known (and most bizarre) example is Moltbook, where a developer told an AI agent running on OpenClaw to build him a Reddit-like platform for AI agents.

The Moltbook homepage.

Less than a week later, Moltbook had more than 1.5 million registered agents that posted more than 100,000 messages to each other. AI agents on the platform soon built their own porn site for robots, and launched a new religion called Crustafarian with a figurehead modeled after a giant lobster. One bot on the forum reportedly found a bug in Moltbook’s code and posted it to an AI agent discussion forum, while other agents came up with and implemented a patch to fix the flaw.

Moltbook’s creator Matt Schlict said on social media that he didn’t write a single line of code for the project.

“I just had a vision for the technical architecture and AI made it a reality,” Schlict said. “We’re in the golden ages. How can we not give AI a place to hang out.”

ATTACKERS LEVEL UP

The flip side of that golden age, of course, is that it enables low-skilled malicious hackers to quickly automate global cyberattacks that would normally require the collaboration of a highly skilled team. In February, Amazon AWS detailed an elaborate attack in which a Russian-speaking threat actor used multiple commercial AI services to compromise more than 600 FortiGate security appliances across at least 55 countries over a five week period.

AWS said the apparently low-skilled hacker used multiple AI services to plan and execute the attack, and to find exposed management ports and weak credentials with single-factor authentication.

“One serves as the primary tool developer, attack planner, and operational assistant,” AWS’s CJ Moses wrote. “A second is used as a supplementary attack planner when the actor needs help pivoting within a specific compromised network. In one observed instance, the actor submitted the complete internal topology of an active victim—IP addresses, hostnames, confirmed credentials, and identified services—and requested a step-by-step plan to compromise additional systems they could not access with their existing tools.”

“This activity is distinguished by the threat actor’s use of multiple commercial GenAI services to implement and scale well-known attack techniques throughout every phase of their operations, despite their limited technical capabilities,” Moses continued. “Notably, when this actor encountered hardened environments or more sophisticated defensive measures, they simply moved on to softer targets rather than persisting, underscoring that their advantage lies in AI-augmented efficiency and scale, not in deeper technical skill.”

For attackers, gaining that initial access or foothold into a target network is typically not the difficult part of the intrusion; the tougher bit involves finding ways to move laterally within the victim’s network and plunder important servers and databases. But experts at Orca Security warn that as organizations come to rely more on AI assistants, those agents potentially offer attackers a simpler way to move laterally inside a victim organization’s network post-compromise — by manipulating the AI agents that already have trusted access and some degree of autonomy within the victim’s network.

“By injecting prompt injections in overlooked fields that are fetched by AI agents, hackers can trick LLMs, abuse Agentic tools, and carry significant security incidents,” Orca’s Roi Nisimi and Saurav Hiremath wrote. “Organizations should now add a third pillar to their defense strategy: limiting AI fragility, the ability of agentic systems to be influenced, misled, or quietly weaponized across workflows. While AI boosts productivity and efficiency, it also creates one of the largest attack surfaces the internet has ever seen.”

BEWARE THE ‘LETHAL TRIFECTA’

This gradual dissolution of the traditional boundaries between data and code is one of the more troubling aspects of the AI era, said James Wilson, enterprise technology editor for the security news show Risky Business. Wilson said far too many OpenClaw users are installing the assistant on their personal devices without first placing any security or isolation boundaries around it, such as running it inside of a virtual machine, on an isolated network, with strict firewall rules dictating what kinds of traffic can go in and out.

“I’m a relatively highly skilled practitioner in the software and network engineering and computery space,” Wilson said. “I know I’m not comfortable using these agents unless I’ve done these things, but I think a lot of people are just spinning this up on their laptop and off it runs.”

One important model for managing risk with AI agents involves a concept dubbed the “lethal trifecta” by Simon Willison, co-creator of the Django Web framework. The lethal trifecta holds that if your system has access to private data, exposure to untrusted content, and a way to communicate externally, then it’s vulnerable to private data being stolen.

Image: simonwillison.net.

“If your agent combines these three features, an attacker can easily trick it into accessing your private data and sending it to the attacker,” Willison warned in a frequently cited blog post from June 2025.

As more companies and their employees begin using AI to vibe code software and applications, the volume of machine-generated code is likely to soon overwhelm any manual security reviews. In recognition of this reality, Anthropic recently debuted Claude Code Security, a beta feature that scans codebases for vulnerabilities and suggests targeted software patches for human review.

The U.S. stock market, which is currently heavily weighted toward seven tech giants that are all-in on AI, reacted swiftly to Anthropic’s announcement, wiping roughly $15 billion in market value from major cybersecurity companies in a single day. Laura Ellis, vice president of data and AI at the security firm Rapid7, said the market’s response reflects the growing role of AI in accelerating software development and improving developer productivity.

“The narrative moved quickly: AI is replacing AppSec,” Ellis wrote in a recent blog post. “AI is automating vulnerability detection. AI will make legacy security tooling redundant. The reality is more nuanced. Claude Code Security is a legitimate signal that AI is reshaping parts of the security landscape. The question is what parts, and what it means for the rest of the stack.”

DVULN founder O’Reilly said AI assistants are likely to become a common fixture in corporate environments — whether or not organizations are prepared to manage the new risks introduced by these tools, he said.

“The robot butlers are useful, they’re not going away and the economics of AI agents make widespread adoption inevitable regardless of the security tradeoffs involved,” O’Reilly wrote. “The question isn’t whether we’ll deploy them – we will – but whether we can adapt our security posture fast enough to survive doing so.”

Krebs on Security – ​Read More