Agentic AI creates management, know-how, and buying issues. Safety leaders have to know the controls that they need to fulfill, the applied sciences that may fulfill them, the place current instruments already present protection, and the place a brand new funding truly fills a niche. Far too typically, we see purchasers conducting that course of in reverse order … attempting to purchase a know-how after which mapping it to controls.
Most of our AEGIS steering periods ultimately attain the identical set of questions: Which applied sciences do we’d like, which controls do they fulfill, and which distributors ought to we study first?
Our newest analysis, Navigate AEGIS Applied sciences To Safe Agentic AI, maps AEGIS controls to applied sciences, performance, and distributors. We did this to provide purchasers a cleaner, simpler path to observe for securing agentic AI.
Map AEGIS Controls To The Tech That Secures Agentic AI
Securing agentic AI doesn’t imply that safety groups must discard their current tech stack and begin over. Most of the capabilities wanted to safe agentic AI exist already in current safety instruments (particularly in the event you work with the massive, acquisitive safety platform distributors).
The issue comes from figuring out the place that current protection ends and what to prioritize. To assist handle this, the report types the record by will need to have now, ought to have subsequent, and specialised and high-assurance use circumstances.
Some acquainted applied sciences now help AI-specific use circumstances. Others present solely a part of the required management. New classes fill gaps created by autonomous brokers, device calls, delegated permissions, mannequin dependencies, and close to instantaneous selections.
A helpful know-how map ought to reply seven questions for every class:
- What the know-how does
- The place it runs
- What it protects
- Which current safety applied sciences share related capabilities
- Which AEGIS controls it helps
- Which NIST AI Threat Administration Framework (RMF) controls it maps to
- Which distributors purchasers can take into account
Our new analysis covers 23 know-how domains throughout the agentic AI stack. These domains embody speedy priorities equivalent to AI runtime safety, AI detection and response, knowledge loss prevention (DLP) for AI, AI safety posture administration, AI id and entry administration, and AI governance, threat, and compliance.
Begin With Management Gaps, Then Work Towards Merchandise
Safety know-how analysis normally begins with a product class. Patrons learn a definition, overview a market, examine distributors, after which attempt to join the class again to an actual management hole. Our report permits groups to reverse that sequence.
This report provides you a sensible determination path by its methodology:
- Determine the lacking or weak AEGIS management.
- Discover the know-how classes that help it.
- Assessment the required performance and deployment location.
- Test for overlap with merchandise already within the surroundings.
- Decide whether or not the group can take in the requirement into an current platform or wants a devoted funding.
- Use the pattern vendor record to start market analysis and analysis.
Right here’s An Instance: AI Runtime Safety
Assume that a corporation identifies gaps in AEGIS controls protecting runtime monitoring, unsafe agent conduct, immediate injection, knowledge exfiltration, or high-risk device actions.
These use circumstances are happy by AI runtime safety.
What the know-how does and what it protects: AI runtime safety screens AI functions and brokers whereas they execute. It collects mannequin and tool-call telemetry, detects exercise equivalent to immediate injection, jailbreaks, knowledge exfiltration, and anomalous actions, after which applies insurance policies to dam, include, redact, restrict, or escalate the exercise.
The place it runs: It could run at an AI gateway, API proxy, utility runtime, agent framework plug-in, sidecar, or one other inline enforcement level.
Current safety applied sciences that share related capabilities: The report reveals the place AI runtime safety overlaps with applied sciences equivalent to AI detection and response, DLP for AI, and AI safety posture administration. That offers safety leaders a possibility to examine their current safety instruments earlier than opening a brand new procurement cycle.
AEGIS and NIST AI RMF alignment: Management mapping helps safety leaders perceive these boundaries earlier than they commit finances.
This control-first view solutions six questions that usually sprawl throughout separate analysis notes, conferences, emails, spreadsheets, slide decks, requests for info, RFPs, and vendor demonstrations:
- What functionality do we’d like?
- The place ought to it run?
- What property and exercise ought to it defend?
- Which management does it fulfill?
- Will we already personal a few of it?
- Which distributors ought to we study?
Use AEGIS As An AI Safety Funding Map
Forrester’s AEGIS framework provides safety leaders a method to outline these guardrails throughout agent conduct, delegated authority, knowledge publicity, device use, mannequin dependencies, and incident response. The following job is changing these guardrails into structure and funding selections. Meaning linking every management to the applied sciences that may implement, monitor, govern, or validate it.
Begin with the management hole. Hint it to the related know-how classes. Test the place current instruments already cowl the requirement. Then resolve whether or not to configure, mix, purchase, change, or wait. Learn the total report for all of the insights and a deep dive into the methodology, applied sciences, and vendor options.
Join With Me
Forrester purchasers with questions associated to this analysis can join with me by an inquiry or steering session.
