AI Security Budget hot talks!

Concern over AI Security has increased, researchers are focusing on Browser Exploitation leveraging AI calling it bragjack.

NEWS

• IANS & Artico surveyed more than 500 CISO about budgets on Cyber Security and it turns out that security budget grew just 5% in 2026 from 4% in 2025
• BragJack
  • A new proof-of-concept attack called BragJack has compromised five major agentic browser environments including Google Chrome, Microsoft Edge, Opera Neon, Perplexity Comet, and Claude.
  • Unlike typical AI attacks, BragJack does not rely on prompt injection or bypassing guardrails; instead, it exploits a fundamental architectural flaw in how browser agents interact with extensions.
  • The vulnerability allows untrusted extensions to hijack communication channels and force the built-in AI agent to execute commands, access files, or leak sensitive data like email correspondence.

Discussion

What zero-day response should be in post-mythos era?

Consider this scenario: “A CVE 2026-001 drops in your feed “unauthenticated RCE, no patch”. You run a version check —> 20 assets match. Management calls … “are we exposed ? what are we doing?”. There are two questions to answer:

1. Are these 20 assets actually exploitable, in my environment ?
2. Would my security controls stop it, right now?

NOTE: patching is off the table, since no patch is available.

1. 8:00 - 8:15 | Day-one Disclosure (No Patch, No Exploit)
2. Audit Exposure Immediately: Run an immediate version check to identify potentially affected assets.
3. Map and Test the Exploit Chain: Since no working exploit exists, map the CVE to its required attack chain (delivery, execution, privilege escalation, process injection, credential access).
4. Simulate Techniques: Execute safe simulations of these specific techniques against your live security controls (NGFW, WAF, Endpoint, EDR, SIEM) to establish actual exploitability per asset.
5. Use your automated pentesting tool to exploit the 20 assets … but there is no public exploit avilable. If you wait for a public exploit, the first working one you see may be the one that hits you.
6. 08:30–08:45 | Control Hardening & Validation
7. Deploy Compensating Controls Immediately: 1. Configure detection rules for the NGFW and SIEM. 2. Apply prevention rules to the WAF. 3. Implement GPO hardening on endpoints and IOA rules on EDR.
8. Staging Patch Tickets: Open parked patch tickets for all affected assets to be executed as soon as vendor patches are released.
9. Re-test the Chain: Re-run the simulated attack chain immediately to verify that controls transition to detected and blocked before a public exploit exists.Idea
10. 12:00–12:30 | Threat Intelligence & Full Campaign Rehearsal
11. Expand Beyond the Vulnerability: As adversary details emerge, treat the CVE as a single link in the broader kill chain (initial access, lateral movement, persistence, exfiltration).
12. Simulate Full Adversary Campaigns: Run end-to-end campaign rehearsals against your live controls to surface collateral security gaps—such as persistence or lateral movement techniques—that pure CVE scanning misses.
13. 16:00–16:30 | Public Exploit Verification
14. Execute Targeted Exploitation: Once a public PoC drops, run automated penetration testing using the live exploit only on assets permitted by policy.
15. Isolate High-Risk/Unpatchable Assets: For assets verified as exploitable where no patch exists, restrict exposure (e.g., place them behind strict WAF rules with access limited to trusted IPs) and escalate patch ticket priority with exploitation proof attached.

Solve this:

How to approach the situation when a new CVE drops (using Mythos or futuristic LLMs) … you’re exposed and … the patch is not available yet.