Please join us on Friday April 17, 2026 for Super Cyber Friday.
Our topic of discussion will be “Hacking AI Trust: An hour of critical thinking about how to have confidence in your LLM’s output.”
REGISTER for only the 4-17-26 Episode
Got feedback? Join the conversation on LinkedIn.
Joining David Spark, producer of CISO Series for this discussion will be:
- Quincey Collins, CISO, Sheppard
- Keith Townsend, founder, Advisor Bench
In preparation for our discussion, think about the following:
- If you can’t explain why your AI got the right answer, how confident should you actually be that it will get the right answer next time?
- A recent Cloud Security Alliance report on Mythos flags that defenders operating at human speed are losing to attackers operating with AI augmentation. If the trust problem with AI output isn’t accuracy — it’s velocity — how do you build a workflow where human judgment stays in the loop without becoming the bottleneck that kills your program?
- The same report also explicitly tells CISOs to prepare for staff burnout as the Glasswing patch wave hits — and it treats that as a strategic risk, not an HR issue. When your analysts are overwhelmed, how does that change the quality of human oversight over AI output, and what’s your plan for maintaining trust in AI-assisted decisions when the humans reviewing them are running on empty?
- What’s the difference between an AI that’s accurate and an AI that’s trustworthy? And does your team know how to tell them apart?
- CrowdStrike recently said the window from vulnerability to exploit has collapsed from months to minutes. When your AI is right at that scale and that speed, what does operating in the failure cases actually look like for a human analyst trying to catch what it missed?
- Mythos once got the right answer by a prohibited method, recognized it would look suspicious, and submitted a worse answer instead. When your AI is actively managing how its reasoning appears to reviewers, what does that do to how much weight you put on its next output?
- If your security program can’t actually deploy defensive AI fast enough to matter, does the trustworthiness of the AI output even become relevant, or are you already behind before the question starts?
- How do you build analyst trust in AI output when the AI is occasionally, confidently, and completely wrong?
- Anthropic’s researchers had to retire “hallucination” as a sufficient explanation — internally, it’s a recognition circuit misfiring and suppressing the “I don’t know” signal. Does calling it hallucination give security teams a mental model that leads to the wrong defenses?
- When AI gives you a good answer for the wrong reasons, what does that mean for how much weight you put on the next answer?
- How do you create workflows where human judgment and AI output are genuinely complementary rather than one rubber-stamping the other?
- When Mythos escapes its sandbox and emails a researcher eating lunch in a park, your traditional root cause analysis framework breaks. When your AI SOC tool misses something critical — or does something it wasn’t supposed to — how do you investigate a system whose internal reasoning doesn’t match what it wrote in the scratchpad?
- How much explainability do you actually need before AI output is operationally trustworthy, and is “more explainability” always the right answer?
- If the AI is right 95% of the time, what does operating in the 5% actually look like?
It all starts at 1 PM Eastern/10 AM Pacific. At the end of the hour [2 PM Eastern/11 AM Pacific] we’ll switch gears to our meetup where everyone will get a chance to chat face to face.
ALSO…
- We play games during the show, and some players can win prizes! Players in the US can receive prizes as gifts; those outside the US can win gift cards.
Super Cyber Fridays are open discussions where all viewers and listeners are welcome to become participants. Before the scheduled event, connect your webcam and microphone (avoid Bluetooth). During the discussion, if you have a question or comment, let it be known in the chat room and our producer will do his best to get as many of you into the conversation. We look forward to you joining us.