Security

SailPoint wrote the rulebook for AI agent tool calls. The platforms bought it instead.

SailPoint published the rulebook for policing AI agent tool calls on October 31. Workday, Cognition and Akamai spent October buying or shipping the same layer, and the December print must show whether anyone pays SailPoint for it.

Vincent JiangVincent Jiang · 2 min read
Share
Mark McClain, CEO and founder of SailPoint
1 / 6Slide 1 of 6
Mark McClain, CEO and founder of SailPoint, which published a blueprint for policing AI agent tool calls on October 31.

The document itself is quiet about money. On October 31 SailPoint−0.10% — SailPoint, down 0.10 percent today, the identity security vendor, published a blueprint for policing AI agents at the MCP layer, the interface where an agent calls a tool such as a database or a business application 1. Before any tool call runs, four things get checked: which agent is asking, which server it targets, what action it wants, and which human started the session. Deny policies, intent classification and session logs follow the call from prompt to attribution. 1

Deployment outran the controls: 5% of APAC enterprises govern agent access

020406080100Run agents unsupervised95Agents in production85Manage agent identities manually68Can detect shadow AI30Early in identity security45Tools governing agent access5
Data
Value
Run agents unsupervised95
Agents in production85
Manage agent identities manually68
Can detect shadow AI30
Early in identity security45
Tools governing agent access5
Share of APAC organizations surveyed in SailPoint's Horizons of Identity Security Report, October 2026. Source: Asia Pacific Security Magazine [2].2

The blueprint cites the hole it fills: 55% of organizations run system-changing AI agents with no identity controls, and researchers have logged 253 MCP server bugs plus six framework bypasses. 1 SailPoint's own APAC research, published days earlier, sizes the customer base: 85% of Asia-Pacific enterprises run agents in production, but only 5% have deployed tools governing what agents can access, 95% run agents unsupervised, and 68% manage non-human identities by hand. 2

The control point is being bought, not licensed

The complication is that platforms are taking that layer rather than paying a specialist for it. On October 13 Workday+0.02% — Workday, up 0.02 percent today agreed to acquire C1, whose Agent Runtime Governance already ships intent-based access controls and enterprise-managed authorization for MCP-connected agents; analysts framed the deal as pressure on SailPoint's identity layer. 3 Workday has 17 agents in production and reports on November 24, the first print carrying the purchase. 4

Cognition~$10B — Cognition, private, latest valuation ~$10B folded the security startup Corridor into Devin Security on October 28, and Akamai's−1.18% — Akamai, down 1.18 percent today Agent Sandbox isolates agent-written code in WebAssembly environments with outbound connections blocked by default. 5 6 The counterpoint: platform bundling does not always beat specialists, and governance spend often starts after the first enforcement action, not before. Regulators from the EU AI Act to Five Eyes guidance already assume an agent's action can be traced to a named person. 1

One customer would settle it

Watch SailPoint's December results for one thing: a named customer paying for agent authorization rather than a design document. Absent that, authorization becomes a payroll feature.

Get Security by email, free

The breaches, flaws and security vendors AI is putting to the test. In our daily email, every weekday.

Loading SAIL

Reader comments

0 comments

    More in Security

    All in this topic →

    Security

    Tenable's Q4 guide leaves out PHP2Shell demand, and CERT-FR flagged a 9.9 flaw in Tenable's own product

    Tenable beat its third quarter and ran nine sessions to a one-year high, but its own fourth-quarter guide prices the AI-and-PHP2Shell demand story at zero, and France's CERT-FR just flagged a 9.9 flaw in the company's identity product.

    Security

    Cellebrite's 154 European police contracts total €69.4 million, and the EU Ombudsman is probing the secrecy

    A Statewatch map counts 154 public contracts worth €69.4 million across Europe, EU funds included, while the Ombudsman probes EU secrecy over the company. CLBT reports Q3 on November 12.

    Security

    1,000 exposed GitLab servers later, the patch clock belongs to a partner

    Shadowserver counts 1,000-plus servers exposed to an actively exploited CVSS 10.0 GitLab bug and 11,300-plus to a second one, after CISA's federal deadline passed. GitLab's answer: a partner, RapidFort, now sells the hardened patch clock.

    Security

    Counterfeit Claude update puts the cleanup bill on enterprise buyers, not on Cursor

    Forged updates to Anthropic's Claude Code sat inside Cursor's official marketplace for about half a day and reached at most 5,000 developers. Five days on, no one has said how code published under Anthropic's name passed the marketplace's checks.

    Security

    CrowdStrike ships a model trained on AI attacks as SentinelOne launches a rival

    Fal.Con Europe opened in Barcelona with the first SafeMind-trained model due in Falcon this week, days after CrowdStrike reported an agent-driven campaign that hunted for machines without its sensor.

    Security

    Rubrik's best quarter met a 10.1% selloff, and 138 insider sales with zero buys

    Rubrik beat and raised, Wedbush and Morgan Stanley priced in the breach wave, and the stock still fell 10.1%. Meanwhile 138 straight insider sales, zero buys, keep printing against the newest bullish coverage.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.