The Mythos CISO Briefing: What I Actually Worked On This Weekend

On Friday afternoon, Gadi Evron asked if I wanted to collaborate on a CISO community document. "Releasing Monday." I said "Sure." (That word has cost me more sleep than any APT.)

Anthropic had announced Mythos on Monday. CSA was running an emergency CISO Zoom on Tuesday. SANS was already building our BugBusters webcast with Ed Skoudis, Joshua Wright, and Chris Elgee for Thursday. Everybody smelled the same thing, everybody was building their own response, and nobody was coordinating.

The entire community was asking the same question: what do we actually DO about this?

Three nights later - Friday, Saturday, Sunday - Gadi Evron, Rich Mogull, and I had a 30-page strategy briefing with 60+ contributors from CSA, SANS, [un]prompted, and the OWASP GenAI Security Project. Contributing authors included Jen Easterly, Bruce Schneier, Chris Inglis, Phil Venables, James Lyne, Heather Adkins, Rob Joyce, Sounil Yu, and a dozen others who all said yes within hours.

The briefing is called "The AI Vulnerability Storm: Building a Mythos-Ready Security Program." It's free. It's a draft. It's live now. And it's the thing I wish existed on Tuesday when my inbox started filling up with "what do we do?"

Read it here.

Listen, Watch, or Quiz Yourself

We built the full NotebookLM resource kit: audio deep dive, cinematic video overview, and three quizzes (board prep, technical team, general knowledge).

Same content, different format. If 30 pages feels like a lot right now, start here.

Lines From the Briefing Worth Stealing

Before I get into what I worked on, here are the lines from this document that I think will stick with people. Use them in your board decks, your team meetings, your Slack channels. They're shareable because they're true.

"The window between discovery and weaponization has collapsed into hours." That's the one-sentence version of why this matters. Tape it to your monitor.

"Security teams are caught in a vice: AI is simultaneously accelerating the volume of vulnerabilities they must respond to and the volume of code their organisations are shipping." Both directions at once. More vulns to fix, more code creating them. (Welcome to the squeeze.)

"Using a coding agent is now easier than using Excel." The number one objection to agent adoption is "my team doesn't have the skills." They do. If they can type, they can start.

"Long-term goals should be considered as a quarter away at most." This one will make program managers flinch. Good. The planning horizons most organizations use were built for a threat landscape that no longer exists.

"Mythos generated 181 working exploits on Firefox where Claude Opus 4.6 succeeded only twice under the same conditions." 181 to 2. That's not incremental improvement. That's a different category of capability.

"This is a normal response to disruptive capability shifts, not a crisis of relevance." For everyone who read the Mythos announcement and felt a pit in their stomach about their career. You're not obsolete. You're adapting. There's a difference.

"We have done this before. Y2K was a systemic threat with a hard deadline, and the industry met it through coordinated, disciplined effort." The closer. And it's true. (Just with less sleep this time.)

The Risk Register

Most of the Mythos commentary this week has been either "we're all going to die" or "this is just marketing." Neither helps a CISO walk into a Monday morning meeting with a plan. So instead, we built a risk register. 13 rows. Grouped by severity. Mapped to OWASP LLM, OWASP Agentic, MITRE ATLAS, and NIST CSF 2.0.

The thing I pushed hardest on was framing every risk as an acceleration of something that already existed, not as a new problem Mythos created. AI-driven vulnerability discovery has been happening for over a year. Mythos made it faster. That distinction matters because it changes how you respond. You're not starting from zero. You're adjusting timelines.

(If I had a dollar for every "unprecedented" I killed in this document, I could fund the next hackathon.)

A few rows worth calling out:

Row 1 (Accelerated Threat Exploitation) includes a point most coverage missed: every patch is now an exploit blueprint. AI accelerates patch-diffing and reverse engineering of fixes. The patch that protects you also teaches attackers where the vulnerability was. (Responsible disclosure just became an attacker tutorial service. You're welcome.)

Row 5 (Insufficient AI Automation) is the one that will make people uncomfortable. It says if your defensive teams aren't using AI agents, they can't match the speed of AI-augmented threats regardless of their technical skill. The asymmetry is cultural, not just technological. (Turns out telling 250 CISOs their teams need to change generates the same reaction as telling 250 CISOs their teams need to change. Who knew?)

Row 12 (Regulatory Exposure) is the one boards need to hear. The EU AI Act hits in August 2026. When AI can find vulnerabilities at accessible cost the standard of "reasonable defensive effort" shifts. Boards will face questions about whether they used available AI tools for scanning, and whether not doing so constitutes negligence.

Row 13 (AI Hype and Confusion) is the one I added because the noise itself is a risk. Teams that dismiss this as hype, or exhaust their attention parsing every hot take, will miss the actual changes they need to react to. (The irony of adding a risk about too much AI content to a document about AI is not lost on me.)

The real problem isn't that Mythos exists. It's that your defensive deployment process was designed for quarterly software releases and regulatory cycles measured in months. (And yes, your CISO and their legal team are both thinking about SNEAKERS right now, even if they've never admit it out loud.)

The Priority Actions Table

11 actions. Ordered by urgency. Each one has a start date, a horizon, and a "What It Means" description specific enough that you can hand it to an engineering lead and they'll know what to do.

The ordering was deliberate and generated significant argument:

PA 1 is "Point Agents at Your Code." Not governance. Not strategy. Point the tools at your code and see what they find. You can start this today with Claude Code Security, Codex Security, OpenAnt from Knostic, or raptor.

PA 6 is the one we added late: "Update Risk Models and Reporting." Because we realized the risk register had a row about outdated risk models but no Priority Action telling CISOs to actually update them. The gap between "you have a problem" and "here's what to do about it" is where most documents fail. We tried not to.

PA 11 is VulnOps. The longest-horizon item. Build a permanent vulnerability operations function, staffed and automated for continuous AI-driven discovery across your entire software estate. This used to be too expensive for most organizations. AI changes that math. (Your quarterly pen test is adorable. The attackers run continuously.)

The Board Briefing Section

This is the section I'd print and hand to a CISO who has 15 minutes to prep for a board meeting. It's organized around the two things CISOs do in front of directors: justify the current program and ask for what comes next.

The strongest line in the paper (Rich wrote it): "The security program this board has funded is what makes the AI strategy viable." That reframes the entire conversation from "security needs more money" to "your AI strategy depends on what you already invested in security." (That sentence alone is worth printing on card stock and sliding under the CFO's door.)

What I'd Do With This Document

If you're a CISO: read the Priority Actions table and the risk register. Skip to the board briefing section if you have a board meeting this month. Print the 10 Questions table and use it to triage where your program actually stands, not where your last audit said it stands.

If you're a practitioner: read the "Human Cost" section. It's honest about burnout, the cultural challenge, and the fact that the practitioners who adapt fastest are the ones who lean into AI tooling rather than viewing it as a threat. Then go point a coding agent at your code. Today.

If you manage security teams: the paper says long-term goals should now be considered a quarter away at most. That changes planning, budgeting, hiring, and how you talk to your board about timelines. Adjust accordingly.

Tune in to BugBusters on Thursday, April 16

If you want to see AI-assisted vulnerability discovery against real code (not slides, not theory, actual terminals and actual bugs), SANS is running BugBusters on Thursday, April 16 at noon ET. Ed Skoudis, Chris Elgee, and Joshua Wright. No registration required.

The briefing will keep being updated. We shipped it now because CISOs need it now, not when it's perfect. (Perfect is the enemy of Monday morning.)