This session demonstrates how quickly and easily enterprise-grade AI security can be applied to your Amazon Bedrock projects. See the new Prisma AIRS AI-security platform in action on AWS: We explore Model Scanning and AI Posture Management; then dive into Runtime Security; and finally examine automated AI Red Teaming and AI Agent Security. Unlock your AI-led innovation - securely.
What this session is about
Playbook
Editorial commentary · what to actually do about this on Monday
The concept
AI security splits into four distinct surfaces: model supply chain (scanning), configuration (posture), inference traffic (runtime), and adversarial probing (red teaming). Each needs different tooling.
Why it matters
Securing AI is not "put it behind a WAF." Prompt injection bypasses traditional perimeters. Models can leak training data. Tool-calling agents have privileges your APIs never granted.
The hard parts
Threat models for agents look different from threat models for APIs. Your existing playbook misses the new vectors.
Playbook moves
(1) Threat-model the agent's tool surface specifically. (2) Separate "data exfil via response" from "command injection via tool call" — they need different controls. (3) Schedule automated red-team runs into the release pipeline.
The surprise
The riskiest agent vulnerability isn't prompt injection — it's *authorisation confusion*. The agent inherits user identity for context but operates with system privileges for execution. Whoever can manipulate the prompt can effectively act as the system. This is the agent-era equivalent of SSRF, and almost no one is testing for it.
---
Independent editorial perspective — not an official AWS or speaker statement. Designed for executives evaluating what to brief their teams on next.
Live updates related to this session LIVE
Sourced via Parallel AI Monitor — continuous web watch on 21 topical streams. Updated .
- onetrust.com high confidence Agent governance & policy gating
OneTrust Unveils Platform and Product Innovations to ...
OneTrust announced new platform capabilities at TrustWeek 2026, including an AI Control Plane that evaluates agent actions at runtime and applies policies to allow, block, or require human approval. It also introduced a headless MCP Gateway experience for AI-native integrations.
- cyberpress.org Agent identity & delegation
RSA Launches Agentic Identity Platform to Secure AI ...
In this September 30 article, IAPP describes an accountability gap in the Model Context Protocol (MCP): its latest release supports enterprise-managed authorization through existing identity providers, but MCP does not provide built-in standards for tracing and logging agent acti
- activepieces.com Decision lineage & audit trails for agents
What Auditors Check When Reviewing AI Action Logs
“What Auditors Check When Reviewing AI Action Logs” focuses on reviewing AI action logs as forensic evidence of automated decision-making, including checking the audit trail to verify model accuracy. It offers a practical audit-review angle on agent decision accountability.
- siliconangle.com Agent infra funding & M&A
Eve Security bags $4.5M in funding to stop malicious AI ...
Eve Security raised $4.5 million, led by Run Ventures with participation from Dreamit Ventures, Blu Ventures, and LiveOak Ventures, bringing its total seed funding to $7.5 million. The company is building a runtime security layer for AI agents with observability, governance, dete
- shattered.io General tech / AI / startup news
TBC Neuron-Derived AI Video Model on AWS [2026]
Valtech launched Agent Factory at DMEXCO 2026, presenting a governed enterprise platform intended to connect and manage fragmented AI agents across organizations.
External links matched to this session via topic relevance. The KB does not endorse third-party content; verify before citing.