1280 Third-Party Products Embed AI Agents Bypassing SSO in Studied Enterprises
Third-party AI agents embedded in existing platforms evade standard controls and expand unauthorized access. Report data and vendor patterns show inherited permissions create unmonitored blast radii. Independent review of connectivity and activity is required beyond vendor assurances.
The 2026 State of Agent Security Report documents agents arriving through platform updates rather than procurement. Salesforce Slack Code grants agents GitHub and production reach via channel membership alone, inheriting permissions without separate review. This matches patterns in prior supply-chain incidents where transitive access expanded blast radius beyond initial configuration screens.
Evidence from contract awards and vendor announcements shows inherited agents now dominate over built or bought variants. Connectivity questions reveal the gap: agents reach data warehouses and ticketing systems through grants never explicitly approved for autonomous use. Official vendor claims of inherited security models conflict with independent observations of silent registration under builder accounts.
Third-party oversight remains insufficient because identity stacks govern only authenticated flows. The four review areas—identity, permissions, connectivity, activity—expose scaffolding risks that model scanning misses. Enterprises face expanding surfaces where agents execute across fixed application edges.
Next steps include mapping transitive grants before Q2 2027 deployments and requiring named human owners for any agent touching production systems.
Salesforce Slack Agent: 30% of production pull requests originate from agents without explicit human approval by Q4 2027
Sources (2)
- [1]Primary Source(https://thehackernews.com/2026/10/the-third-party-agent-problem-why.html)
- [2]Supporting Source(https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-10)