I work at the intersection of cyber defense, AI safety, and emerging risk, where the threats are converging and the playbooks are still being written.

My career has taken me from the Marine Corps to NSA, U.S. Cyber Command, the U.S. House of Representatives, Mandiant, and Netflix. Across every role, one throughline: understanding adversaries, building high-trust teams, and translating complex problems into strategies leaders can act on.

Today, I advise organizations, governments, and research institutions on AI governance, agentic threat preparedness, and unifying safety and security into a single discipline. I contribute to global standards efforts including the EU AI Act (via OWASP AI) and lead executive education on cybersecurity and AI strategy at SANS.

If you’re navigating the future of AI risk or security strategy, let’s connect.

Chris Cochran

This room was electric.

Just wrapped my Agentic Identity talk / BoF at [un]prompted, and honestly, having some of the best minds in AI security in one room was a highlight of the conference. We didn’t keep it high-level, we went deep into the architecture of the future.

Several of the folks in the room did research that started my own journey, or are actively tackling this same challenge:
Ken Adler, John Yeoh, Toby Kohlenberg He/Him, Charles Nwatu, Rob T. Lee, and Olivia Gallucci, just to name a few.

The conversation pushed into some big questions:

• The Agent Era: What does identity look like when the “user” isn’t human but can sometimesact like one?
• Authorization: How do we solve who can do what in a generative world?
• The Infrastructure Question: Centralized platforms, blockchain experiments, or an inevitable hybrid?

The questions from the room were next-level.

Huge thanks to Gadi Evron, Sounil Yu, and the entire [un]prompted team for creating space for these conversations.

I’ll share a deeper recap (and a few emerging answers) soon.

Field CISO & VP, AI Security @ SANS | Bridging Cyber Defense and Frontier AI Risk | Ex-NSA, Netflix, Mandiant | US Marine Veteran

No alternative text description for this image

Programs like this can change a career forever.

Honored to be a small part of this amazing program.

https://lnkd.in/ghuY7wiG

Women in CyberSecurity (WiCyS) SANS Institute Foster Nethercott Mari DeGrazia Maxwell Shuftan Tyler Corcoran Kajal Shelat

Open article: Women in CyberSecurity (WiCyS) Launches AI Security Accelerator Program for Maryland Members by finance.yahoo.com

Women in CyberSecurity (WiCyS) Launches AI Security Accelerator Program for Maryland Members

Excited to be in San Francisco for the first-ever [un]prompted Conference.

It’s encouraging to see a room full of practitioners leaning into the evolving discipline of AI security, particularly around agentic systems, identity, and workforce readiness. Not long ago, this wasn’t even a defined field. That’s wild.

If you’re here and interested in thoughtful conversations about securing what’s next, including what we’re seeing globally and how the SANS Institute is supporting the community, I’d welcome the exchange.

Strong work by Gadi Evron, Sounil Yu, and the team bringing this together.

A few years ago, I was done with security. Burned out. Two decades of the same circuit, same panels, same predictions. The people closest to me knew it.

Then AI pulled me back from the dead.

I just spent a week at [un]prompted in San Francisco, and I walked away changed. Not inspired in the vague, conference-high, shallow influencer, kind of way- changed in the "I was wrong about seven things and I'm all in" kind of way.

Wrong about the sidelines. Wrong about the hype. Wrong about whether anyone else saw what was coming.

In my latest article, I break down all seven, plus standout work from folks like Nicholas Carlini at Anthropic, Heather Adkins and John "Four" Flynn at Google, Olivia Gallucci at Datadog, and Scott Behrens and Justice Cassel from Netflix security team, all building the future of AI security in real time.

I even had a bit of a surprise guest spot.

This is our Industrial Revolution, folks. And the rules are still being written.

Check out the seven things I was wrong about below.

Gadi Evron Sounil Yu Ida Vass

I'm a student of history, business, and technology. Always have been. For years, I looked back at the Industrial Revolution, at the dawn of the intern

Open article: 7 Things I Was Proven Wrong About AI Security at the [un]prompted Conference by Chris Cochran • 7 min read

7 Things I Was Proven Wrong About AI Security at the [un]prompted Conference

I’m honored to join Pentera’s AI Security Exposure Survey 2026 Webinar alongside Ryan Dorey, Director of Technical Advisors at Pentera, where we’ll unpack the real deal from the benchmark data revealing how organizations are navigating the security challenges of rapid AI adoption. 

What we’re seeing in the data is pretty straightforward, but some insights will surprise you:

Legacy controls are being stretched to cover AI-driven workflows.
Teams are being asked to secure systems they didn’t design and may not fully understand.
And in many cases, organizations don’t have a clear way to validate AI risk before an attacker does it for them.

In this session, we’re going to have a very practical conversation about that.

We’ll cover:

• How the benchmark was built and what 300 CISOs are actually saying
• Real examples of how AI security gaps become exploitable exposure
• Where adversarial testing fits in, across data, identity, and infrastructure

It’s about closing the gap between AI adoption and real-world security validation.

Join us by registering with the link in the comments.

No industry faces the complexity of AI adoption quite like healthcare. Getting it wrong is measured in revenue but more importantly, it’s measured in outcomes, in trust, and in lives.

That’s what makes the conversations happening right now so critical.

What gave me genuine optimism was seeing security teams and IT leaders speaking with one voice, not just asking “How do we protect the organization from AI threats?” but going further. 

Acknowledging that AI carries extraordinary promise and meaningful risk, and that moving forward demands intentionality above all else.
The next 12–18 months will be defining. 

Organizations that build governance frameworks, align stakeholders, and treat security as a design principle will earn lasting trust. Those that move recklessly won’t get a second chance.

I’m encouraged by an industry leaning in with both urgency and discipline. That balance is rare. It’s also exactly what this moment requires.
Thank you to SANS Institute, HIMSS, and the healthcare community for creating a space where these conversations can happen with the seriousness they deserve.

The work continues.​​​​​​​​​​​​​​​​

It was great to see the masterful Chris Wilkes work his magic in another Crisis Exercise. Amanda Burdey, GISF, GSTRT and I got to do a bit of acting in support. 

It was also a pleasant surprise to see Maria Graham. I still am and will always be jealous of your friend group. So cool to see.

#HIMSS26

Want to see how AI can automate absolutely everything in the SOC?

Then this is not the talk for you.

On March 18th, I’m joining my dear friend Nathan Burke from 7AI for a SANS Institute webcast “Don’t Trust AI – Verify It: A Practical Discussion About AI in the SOC”  and we’re not here to sugarcoat anything.

We’ll be getting into:
• Why agentic AI systems are introducing risks most orgs aren’t ready for
• What “verify, don’t trust” actually looks like in practice inside the SOC

If you’re leading security strategy in an AI-first world, this conversation is one to check out. 

No fluff, no vendor pitches just two crazy tech guys with slightly too much energy, who care deeply about getting this right. 

Get your spot in the comments below.

If you know Kate Marshall, you know she is someone who legitimately wants to give value to everyone in her atmosphere. 

This book is pure value for those looking to bring AI into their personal or professional lives.

I’m proud of Kate, not only for this masterclass of a book, but for stepping into to the arena and arriving like a force to be reckoned with.

Grab your copy here: https://a.co/d/02rpuPWV

Great to be back with our special operations community this week at SOFWERX for C2DS. We focused on the intersection of agentic security and prompt engineering, specifically how these technologies can be weaponized as tactical force multipliers for the mission.

As the threat evolves, ensuring our elite operators have a competitive edge is paramount. It was a privilege to contribute to the mission and spend time with the professionals dedicated to keeping this country safe.

Thanks for making this week awesome Kerry Rock, Luke Leviner, and Foster Nethercott. 

Semper Fi.

Last week I sat down with 300 CISOs worth of data and one question kept echoing in my head: How is it possible that 100% of enterprises have adopted AI, and zero percent have full visibility into it?

I hosted a live session last week with Ryan Dorey from Pentera, digging into their AI Security & Exposure Benchmark 2026. What came out of it was one of the most honest, unfiltered conversations I've had about where enterprise AI security actually stands right now.

A few things that stopped me cold:

→ Only 11% have purpose-built AI security tooling, the rest are stretching legacy controls to cover attack surfaces they were never designed for
→ 50% of CISOs cite lack of internal expertise as their # 1 barrier, not budget
→ Shadow AI isn't some edge case. It's the default condition in most enterprises right now

And the one that always keeps me up at night: we are in the calm before the storm.

Autonomous agents that don't sleep. Attack paths that are 80-90% automated. An 11-year-old with access to a sophisticated offensive agent isn't science fiction, my friends.

I wrote up a bit from the session, the data, the real-world attack scenario we walked through, and what I'd tell every security leader to focus on today.

Link to the full article. 👇

What's the biggest AI security challenge you're navigating right now? I'd genuinely love to hear it.

Last week I had the privilege of joining Ryan Dorey from Pentera to dig into their freshly released AI Security & Exposure Benchmark 2026 - a survey o

Open article: The 100/0 Problem: Every Enterprise Has AI. None Have Full Visibility Into It. by Chris Cochran • 6 min read

The 100/0 Problem: Every Enterprise Has AI. None Have Full Visibility Into It.

I was fortunate enough to contribute to this CSO Online piece on how CISOs are rethinking data protection in the AI era.

The core challenge I keep coming back to: AI-driven data exposure doesn't look or feel like exfiltration. Employees paste source code and customer data into consumer models, thinking they're being productive,  not realizing the risk. Add expanding data sovereignty rules, AI-specific regulatory guidance, and post-quantum encryption on the horizon, and you understand why this has become a board-level conversation.

Shoutout to the other security leaders and practitioners featured in the article by Mary K. Pratt, lots of practical thinking in here worth your time.

Link in the comments.

It was an honor to hang out with my buddy Christopher Luft to chat all things AI in security.

Thanks LimaCharlie and SANS Institute for facilitating the conversation.

AI skill gaps are a real conversation right now, and Chris Cochran, Field CISO and VP of AI Security at SANS Institute, breaks it down into three practical buckets for defenders who want to stay ahead.

Start by figuring out what you can offload to AI: summarization, enrichment, repetitive tasks. Save the deterministic decisions for humans.

Then learn how to secure AI itself:

> Agent-to-agent communications
> MCP server security
> OWASP Top 10 for agents

Finally, understand governance. Not just the technical side, but what your company is actually trying to do with AI. Security practitioners who can enable the business, not just protect it, become irreplaceable.

Listen to the full episode: https://lc.pub/4s7iL4V

#cybersecurity #ai #podcast

I’m incredibly proud of the work we’ve put into the latest OWASP AIVSS Project update. Still pinching myself about being involved with some of the most respected leaders in our industry.

OWASP AIVSS Project 
OWASP AI Exchange 
SANS Institute

The AI Exchange is proud to co-publish this magnificent OWASP AIVSS Project update. Credits to all experts and for the OWASP AI Exchange: Behnaz Karimi, Chris Cochran, Miracle Owolabi, with supporting: Alex Leung and Srajan Gupta.

#ai #aisecurity

☁️ Francesco ☁️ Cipollone

Reduce risk - focus on vulnerabilities that matter - Contextual ASPM - CEO & Founder - Phoenix security - 🏃‍♂️ Runner - ❤️ Application Security Cloud Security | 40 under 40 | CSA UK Board | CSCP Podcast Host

As the CEO and Founder of Phoenix Security, I lead efforts to revolutionize application security through actionable Application Security Posture Management (ASPM), leveraging machine learning and threat intelligence to accelerate vulnerability resolution by up to 40%. With over four years of experience at the helm, our team empowers organizations to quantify risk reduction and demonstrate the ROI of software security initiatives.  

Additionally, as a faculty member at IANS, I specialize in application security, cloud security, and vulnerability management, providing expert insights and education to elevate organizational security postures. Passionate about fostering innovation in cybersecurity, I collaborate with teams and leaders across industries to create measurable impact while maintaining a commitment to empowering secure engineering environments.

CEO & Founder - Diversity Champion

Working on the future of Application Security Vulnerability Management powered by machine learning, and enriched by threat feed. 
Accelerate the detection and resolution of vulnerability up to 30-40% faster. 
Calculate RoI of security in software and quantify the risk reduction of software security

I tought we were arguing about claude code, and sandwarm appeard inside the repo …

Everyone is busy talking about securing AI coding assistants.

Meanwhile, SANDWORM_MODE is doing something far more practical:
infecting npm, stealing tokens, rewriting repos, and quietly looping through CI.

This is a Shai Hulud style supply chain worm. One typo. One install. Then:

TLDR

- Typosquatted npm packages execute on import
- Steal npm, GitHub, CI, cloud, SSH and LLM API keys
- Modify package.json and lockfiles
- Inject malicious GitHub Actions workflows
- Persist via git config --global init.templateDir
- Drop a rogue MCP server into AI assistant configs

This is not a single host issue. It is repo propagation plus CI amplification.

Exact malicious versions
claud-code@0.2.1, cloude-code@0.2.1, cloude@0.3.0, crypto-locale@1.0.0, crypto-reader-info@1.0.0, detect-cache@1.0.0, format-defaults@1.0.0, hardhta@1.0.0, locale-loader-pro@1.0.0, naniod@1.0.0, node-native-bridge@1.0.0, opencraw@2026.2.17, parse-compat@1.0.0, rimarf@1.0.0, scan-store@1.0.0, secp256@1.0.0, suport-color@1.0.1, veim@2.46.2, yarsg@18.0.1
Watchlist: ethres, iru-caches, iruchache, uudi

Why this is different from earlier Shai Hulud variants

Variant 1, 2 and 3 focused on npm plus credential harvesting plus repo spread.

SANDWORM_MODE escalates:

Built for changeability and mutation
Uses GitHub Actions as a propagation engine
Actively poisons AI tooling via MCP injection

This is dev laptop + CI + assistant compromise in one chain.

What to do immediately

Hunt and remove those exact package versions everywhere
Rotate GitHub, npm, CI, cloud, SSH and LLM keys if installed
Audit .github/workflows for unexpected pull_request_target usage
Check git config --global init.templateDir
Check assistant configs for unexpected mcpServers entries

Quick detection option

Open source scanner:
in comments

Run:
python3 https://lnkd.in/enDeHJ7Y your_repo --enable-phoenix --output scan-report.txt
Clean result: no package hits, no workflow implants
Bad result: treat as credential exposure + repo integrity incident

We can harden Claude Code all we want.
Thanks to Socket for the original disclosure, we better together when the intel is distributed fast 

If your npm graph is poisoned, your CI becomes the attacker’s automation layer.
Full analysis in the comment section

#malware #sandworm

We going to see the new generation of hackers here :) script kiddies now vibe hackers will emerge from here. 

I Kew and embraced llm from the first baby steps but the extent those tool have fundamentally change the industry is mind blowing. 

When people called it bubble and dotcom moment I tought. There is real money and fundamental application in this. 

Welcome to the era of ai everything 

#ai #security #vibecoding

Claude code has killed sast, cloud code has killed the whole cybersecurity market… defence in depth still matters!

Claude's code security is a great skill and a one-click assessment for small apps. I used the same tool in Lovable, and it is great for a quick fix and assessment. 

At enterprise scale? It's good, maybe for a fix or two, but can you practically measure the issue and handle the complexity window?

Complexity is not “more files”. It’s more trust boundaries, more shared services, more IAM edges, more release trains, more compensating controls, more exceptions. A repo-only view will miss why something is safe in one place and catastrophic in another.

Check out Pramod Gosavi post has some good take on the whole market 

I wanted to make 3 points

1) Vibe coding vs enterprise reality

Claude Code can move fast when the problem is local and the blast radius is small. Enterprises live in distributed systems and policy constraints.
	•	Multi-tenant boundaries, data residency, regulated workflows
	•	Dependency graphs with pinned versions, internal forks, and signed artifacts
	•	Shared libraries where a “simple fix” becomes a platform change
Low friction is a feature. Without architecture and guardrails, it becomes a multiplier for unsafe assumptions.

2) The attacker gets the same leverage

The uncomfortable bit: the same reasoning used to locate and patch issues can be used to locate and exploit them.
	•	Faster discovery across large codebases
	•	Better chaining across weak links (auth edge cases, unsafe deserialization, SSRF paths)
	•	More rapid weaponisation once a pattern is found
Security teams cannot treat AI scanning as a nice-to-have. It is now table stakes because the opposing side has similar tooling incentives.

3) The real pull is the enterprise remediation loop

AI SAST is useful. The enterprise win is the full review and remediation cycle:
	•	A consistent workflow for triage, ownership, and prioritisation
	•	Policies that translate findings into enforceable defaults
	•	Fixes that land safely across versions, containers, and environments
This is where AI has more pull: not just finding issues, but compressing the cycle from “finding” to “verified fix in production” with traceability.

My take: Claude Code makes scanning and patch suggestions cheaper. The differentiator is whether your organisation can turn that into a governed remediation engine, not a token-burning firehose.

If you’re rolling this out, the hard questions are simple:
	•	Who owns fixes by service and team
	•	What rules prevent unsafe refactors
	•	How you validate impact across code, dependencies, container builds, and runtime exposure

#cloudecode #appsec #enterprisecoding

taplio

The Battle

Chris Cochran

☁️ Francesco ☁️ Cipollone

Followers

Posts

Engagements

Likes

Performance ScoreChris Cochran's LinkedIn Performance

Performance Score☁️ Francesco ☁️ Cipollone's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Chris Cochran

☁️ Francesco ☁️ Cipollone

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Chris Cochran's LinkedIn Performance

Performance Score☁️ Francesco ☁️ Cipollone's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreChris Cochran's LinkedIn Performance