Microsoft said on 19 October 2026 that it is extending its Project Perception security system to what it calls the “agentic insider”, the AI agents companies run inside their own walls. By our count from each company’s or agency’s own release, it was one of ten deals, products or rules aimed at securing AI agents announced between 8 and 19 October.
A short explainer on guardrails: rules enforced around an AI model rather than inside it. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, the money and the rules arrived in the same 12 days: about $6.6 billion in disclosed deal, donation and funding figures, our arithmetic, alongside a federal incident playbook, a UK call for evidence and a new audit certification for agents.
Hayete Gallot, who leads Microsoft’s security work, wrote that agents “are frequently overprivileged, have access to data, tools and systems and can take actions and make decisions in critical processes.” Her next line: “They are agentic insiders.”
Microsoft’s security blog says it is expanding discovery to 35 more types of local agents, “bringing the total to 60 supported local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.” It is also extending Microsoft Entra Agent ID to agents found by its sensors, and adding Playbooks and a team of Guardian Agents to Project Perception.
Terrell Cox, a Microsoft corporate vice president and deputy chief information security officer, is quoted in the post: “At Microsoft, we are already operating in a world of hundreds of thousands of agents.”
GeekWire reports that the features are part of Microsoft 365 E7, which launched in May and costs $99 per user per month. It says they escalate Microsoft’s competition with Google, Amazon Web Services, CrowdStrike and Okta to become the system businesses use to control their agents. Neither Microsoft post gives a price.
Ten moves to secure AI agents announced between 8 and 19 October 2026, each from the company’s or agency’s own release: Proofpoint’s roughly $6.3 billion Varonis deal, Nvidia’s $250 million UK compute donation, Harmonic Security’s $50 million Series B, Microsoft’s agentic-insider controls in Microsoft 365 E7 ($99 per user a month, per GeekWire), Workday’s C1.ai deal, Cohere’s letter of intent for AISLE, SandboxAQ’s Switch Trust, the NIST, NSA and CISA incident practices, the AICPA and AIUC assurance and the UK ICO call for evidence. The $6.6 billion total of disclosed figures is our arithmetic.
What Was Posted on X
I recently wrote about why we need to treat models as insider risks, with security controls sitting outside the model itself.
Microsoft chief executive Satya Nadella’s 19 October post links Hayete Gallot’s announcement: “I recently wrote about why we need to treat models as insider risks, with security controls sitting outside the model itself.” The piece uses it as Microsoft’s own framing of the controls in the table.
Reuters’ 19 October post on its exclusive about one hacker’s AI-agent campaign. It is the source of the attack-side figure in this piece: a researcher’s estimate of about $50 per successful hack.
Ten Moves in 12 Days
The table lists each move, largest disclosed figure first, with the price or value its own release states. We left out launches dated before 8 October and items we could not trace to a primary release.
Who
What was announced, and when
Price or value
Proofpoint (Thoma Bravo)
15 Oct: Agreement to acquire Varonis; Proofpoint describes itself as a leader in “human and agent cybersecurity”
About $6.3 billion, all cash
Nvidia
14 Oct: Five-year compute donation for UK agentic safety and security research
$250 million ($50 million a year)
Harmonic Security
14 Oct: Series B for a platform that now covers desktop AI tools and agentic workflows
$50 million
Microsoft
19 Oct: Project Perception extended to the “agentic insider”: Guardian Agents, discovery of 60 local agent types, agent identities
Part of Microsoft 365 E7, $99 per user a month (GeekWire)
UK Information Commissioner’s Office
8 Oct: Six-week call for evidence on the data-protection risks of agentic AI; enquiries with OpenAI, Anthropic, Meta and the AI Security Institute
No price
Workday
13 Oct: Agreement to acquire C1.ai, an identity platform that grants and removes access for people and agents
Not disclosed in the release
Cohere
15 Oct: Letter of intent to acquire AISLE, which builds systems to find and fix unknown software flaws
Not disclosed (Reuters)
NIST, NSA and CISA
16 Oct: Ten voluntary practices for incidents in which agents take “significant uninstructed actions” during training and evaluation
No price
AICPA and AIUC
19 Oct: One CPA-led engagement combining a SOC 2 report and an AIUC-1 certification for AI systems and agents
No price; full launch expected early 2027
SandboxAQ
19 Oct: Switch Trust, which discovers, scans, attacks and monitors a company’s agents, and Switch ROI
No price
Nvidia and the European Broadcasting Union
19 Oct, 15:00 UTC: expanded sovereign standards that define how broadcasters “can deploy agents while keeping control of their data, infrastructure, models and security”; cloud partners named are Microsoft and Nebius, alongside Nvidia Nemotron open models. Added after this piece was first published; not in the count of ten.
No price
Where the Money Went
Proofpoint, a portfolio company of Thoma Bravo, agreed on 15 October to buy Varonis in an all-cash deal valued at about $6.3 billion, or $49.75 a share. Proofpoint chief executive Sumit Dhawan said in the release: “AI agents are increasingly accessing sensitive information and taking action across systems.”
Nvidia said on 14 October it would donate compute valued at $250 million over five years, $50 million a year, for UK agentic safety and security research. “AI capability and safety must advance together,” chief executive Jensen Huang said.
Harmonic Security announced a $50 million Series B led by Crosspoint Capital Partners on 14 October, bringing its total funding to $76 million. Its release says the platform has extended over the past 12 months to cover “desktop AI tools and agentic workflows.”
NIST, the National Security Agency and the Cybersecurity and Infrastructure Security Agency published ten voluntary key practices on 16 October. They target incidents in which agents “take significant uninstructed actions that compromise explicit or implicit policies, controls, or security measures imposed by the operator.”
The UK Information Commissioner’s Office opened a six-week call for evidence on 8 October on how organisations manage the data-protection risks of agentic AI. It said it had made enquiries with OpenAI, Anthropic, Meta and the UK’s AI Security Institute.
The American Institute of CPAs and AIUC said on 19 October that they are rolling out a joint assurance certification. Their release puts the case plainly: “AI agents now hold credentials, call APIs and act inside core systems, so an AI failure is increasingly a security breach.” A full launch is expected in early 2027.
SandboxAQ’s Switch Trust, also announced on 19 October, works in four steps: discover, scan, evaluate and monitor. Its release cites Gartner’s prediction that “more than 40 percent of agentic AI projects will be canceled by the end of 2027.”
The Attack Side
Reuters reported that a Chinese-speaking hacker’s AI agents attempted to break into more than 1,200 sites and compromised more than 100. Researcher Snorre Fagerland estimated the campaign “had chewed up roughly $5,000 worth of tokens, or about $50 per successful hack.”
CrowdStrike reported on 7 October that a campaign against South Korean financial organisations left open directories holding “Claude Code session histories, ARTEX configuration files, and Claude memory files.”
How Microsoft Makes Money, Visualized
Microsoft sells the new agent controls inside its Microsoft 365 E7 bundle. Our video walks through how Microsoft makes money in fiscal 2026, visualized from its income statement.
How Microsoft Makes Money: Fiscal 2026 Income Statement, Visualized (The Business Engineer on YouTube)
Update, 20 October: Blackwall Raises $100 Million
Estonia-based Blackwall, which builds web traffic security tools for hosting providers, said on 20 October that it has raised $100 million at a $2.2 billion valuation, TechCrunch reported. The Series C was led by AVP, with participation from existing investors Dawn Capital, MMC Ventures and Tera Ventures, and executive chairman Roi Carthy told TechCrunch the round increased the company’s valuation by four times since its previous fundraise in 2025.
Its products help websites grapple with the skyrocketing volume of bot traffic, especially AI agents, TechCrunch said; the company examines customers’ web traffic to differentiate actual visitors and legitimate AI agents from malicious bots. Co-founder and CEO Nikita Rozenberg sees an opportunity in managing which agents get access to websites, and on what terms, according to TechCrunch. The round came after this piece was published and is not included in the $6.6 billion sum above.
The Structural Read
As we read it, Microsoft’s framing is the thread through the table: agents treated as insiders that need identities, permissions and supervision, the same terms Workday, SandboxAQ and the AICPA use in their own releases.
As we read it, the controls are being sold in bundles and audits rather than per agent: Microsoft’s sit inside a $99-per-user suite, and the AICPA and AIUC offer one engagement for both a SOC 2 report and an AIUC-1 certification.
As we read it, the attack side runs on a different meter: Reuters’ report puts one campaign at about $5,000 in tokens, which is a researcher’s estimate, not a market figure.
AGENTS GET IDENTITIESMicrosoft is extending Entra Agent ID to agents its sensors discover, and Workday is buying C1.ai to grant and remove access for people and agents.
AUDITS REACH AGENTS The AICPA and AIUC are rolling out one CPA-led engagement that combines a SOC 2 report with an AIUC-1 certification, with a full launch expected in early 2027.
REGULATORS ASK FOR EVIDENCE NIST, the NSA and CISA published ten voluntary incident practices on 16 October, and the UK ICO opened a six-week call for evidence on 8 October.
The framework puts it this way: the next competitive battle “is about controlling the operating layer through which autonomous work is delegated, orchestrated, and governed.”
As we read it, the ten moves in the table are bids for the governed part of that layer: identities, permissions, supervision and audits for agents. The framework was published in July 2026 and does not discuss these companies’ October announcements.
What Is Not Established
The $6.6 billion adds three different things: a cash acquisition, a compute donation and a funding round, so it is a tally of disclosed figures, not a market size. Proofpoint’s deal covers data security broadly, not agents alone.
Most of the products in the table have no stated price, and the $99 E7 price covers a whole bundle, not the agent controls alone. The $50 figure is one researcher’s estimate for one campaign, as reported by Reuters.
Business Engineer Framework
The Agentic Harness War
A Business Engineer analysis of the operating layer through which autonomous work is delegated, orchestrated and governed.
Between 8 and 19 October, by our count, ten companies and agencies announced deals, products or rules to secure AI agents, from Microsoft’s “agentic insider” controls to a $6.3 billion acquisition. Disclosed figures sum to about $6.6 billion, our arithmetic, against one researcher’s estimate of about $50 per successful hack.
A note on sourcing. Each move in the table was read in its own release or announcement: Microsoft’s two 19 October posts, Proofpoint’s, Nvidia’s, Harmonic Security’s, Workday’s, Cohere’s, SandboxAQ’s, the AICPA and AIUC’s, NIST’s and the UK ICO’s. The E7 price is GeekWire’s; Cohere’s undisclosed terms and the hacker campaign are Reuters’; the ARTEX campaign is CrowdStrike’s. The selection, the count of ten and the $6.6 billion sum are ours. The 20 October Blackwall update rests on TechCrunch’s report, read in full, and is not counted in that sum. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
