Secure Talk Podcast

The trillion-dollar AI story is about autonomy. The real value — in fraud detection, security triage, and risk scoring — is hiding in the AI almost nobody is talking about.

Most machine learning models built to be deployed never get deployed. Meanwhile, budgets pour into generative AI pilots that dazzle in the demo and stall in production. Data scientist and author Eric Siegel estimates the real value of today's AI may be closer to 5% of what current valuations imply- still enormous, but a lot of disillusionment waiting to happen.

If you lead security, GRC, compliance, or data teams — or you've been told to "do AI" without a clear business case — this conversation is your reality check.

📖 Eric's book: The AI Playbook: Mastering the Rare Art of Machine Learning Deployment. 

Chapters

00:00 Predictive and generative AI as complementary tools

02:33 What predictive AI does that generative AI does not

06:32 How hype around intelligence and agents creates scope creep

09:16 Risk scores and human review for more reliable automation

11:22 Why model accuracy is not the same as business value

13:30 BizML’s six-step path from project idea to deployment

16:08 The second wow: proving operational impact

19:20 AGI claims, uncertainty, and the limits of prediction

23:14 AI investment, business sustainability, and the bubble

27:08 Why predictive and generative AI work better together

31:35 Ethics, protected classes, and machine bias

37:20 Why hiring models cannot repair workplace culture

39:08 How to interpret probabilities without mistaking them for certainty

SOURCES & REFERENCES

• The AI Playbook by Eric Siegel → https://mitpress.mit.edu/author/eric-siegel-39891/

• Gooder AI (Eric's startup) → https://gooder.ai

• Eric's op-eds on discriminatory models → https://civilrightsdata.com

• Cory Doctorow, coiner of "enshittification" → https://pluralistic.net

• ProPublica, "Machine Bias" → https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

• EEOC guidance on AI in hiring → https://www.eeoc.gov/sites/default/files/2024-04/20240429_Employment%20Discrimination%20and%20AI%20for%20Workers.pdf

• Nate Silver's 2016 election forecast → https://www.youtube.com/watch?v=nVNEz0W-Ud4

THE STORY

▶ The first "wow" is easy. The second is rare.

AI pilots knock everyone's socks off — and then die before deployment. Siegel argues that teams obsess over technical benchmarks and lose sight of business outcomes: "It's like being more excited about the rocket science than the launch of the rocket."

▶ AGI and the autonomy sales pitch

Using a chatbot with a human in the loop is valuable. But that's not what's being sold. "AGI is, by definition, the ultimate autonomous story," Siegel says — and that's where the hype outruns the technology.

▶ The mirror problem

Strip out a protected-class column and a model can still discriminate through proxies. Siegel revisits ProPublica's recidivism investigation, which found higher false-positive rates for Black defendants. Justin shares how a hiring model exposed a company quietly weeding curiosity out of its own culture.

🔐 SecureTalk is brought to you by Strike Graph — compliance and security that scales → https://strikegraph.com

ABOUT

Guest: Eric Siegel is a 35-year machine learning veteran, a former Columbia University professor, founder of Machine Learning Week, co-founder of Gooder AI, and author of The AI Playbook and Predictive Analytics.

Host: Justin Beals is the founder and CEO of Strike Graph and a career builder of machine learning products in education, hiring, and security.

📄 Full transcript: https://www.strikegraph.com/secure-talk-podcast/ai-is-worth-5-of-its-hype-agi-myths-machine-bias-predictive-ais-untapped-value
🔔 Subscribe for weekly conversations on cybersecurity, compliance, and the tech reshaping risk.

#AIhype #PredictiveAI #MachineLearning #Cybersecurity #AIgovernance

The internet was built for a handful of universities, with no security where it seemed unnecessary. We’re now making the same mistake with AI, only faster.

The Problem: 

Once a technology spreads, it’s “humpty dumpty.” You can’t put security back in. Brian Cute has watched this happen from the inside: at a submarine cable carrier during the dot-com crash, at Network Solutions when it was the only domain seller on Earth, and inside ICANN. This Cybersecurity Awareness Month, he explains why the rush to adopt AI is repeating the internet’s original design mistake, and what history says we should do instead.

Who is this for: 

Security architects, GRC and compliance professionals, policy watchers, and technology leaders adopting GenAI and agentic AI without a clear risk model.

Summary: 

The Original Sin

Vint Cerf and the early internet’s designers built for scale within a closed group of universities, and some security features were left out. When the network went global, criminals found those gaps. “Getting everybody to adopt a security fix is really, really hard, next to impossible,” Brian says. Everyone rushing to deploy AI, he warns, is watching “a similar movie, yet again.”

Take the Word “Governance” Away

ICANN worked because the DNS is a scarce, single-operator namespace. AI has none of those attributes. Brian argues the real work is building collaboration mechanisms: a defined problem, a shared outcome, and clear roles for government, industry, civil society, and engineers. Justin adds that existing privacy, fraud, and IP law already covers much of what’s needed.

Bring a Tool, Then Get Out of the Way

Brian’s lesson from rural communities: “All of the knowledge that is required to address these problems already exists.” Outsiders should bring a tool that’s fit for purpose and let the community adapt it.

📩 Subscribe to Secure Talk:

🔐 Brought to you by Strike Graph: www.strikegraph.com

Chapters

00:00 Introduction and guest background

04:22 Starting the conversation on internet evolution

09:27 Lessons from the domain name system and security challenges

12:27 AI's impact on cybercrime and automation

17:30 Scaling cybersecurity tools for small organizations

21:47 Measuring security impact and behavior change

25:35 AI as a tool for protection and learning

36:08 Governance of AI and internet infrastructure

40:07 Collaborative approaches to AI governance

43:58 Closing remarks and future outlook

📚 SOURCES

Global Cyber Alliance: https://globalcyberalliance.org

IETF: https://www.ietf.org

Microsoft Digital Defense Report 2025: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Microsoft-Digital-Defense-Report-2025.pdf

CISA Cybersecurity Awareness Month: https://www.cisa.gov/cybersecurity-awareness-month

ICANN: https://www.icann.org

Model Context Protocol: https://modelcontextprotocol.io

ABOUT THE GUEST

Brian Cute is CEO of the Global Cyber Alliance and former CEO of Public Interest Registry, operator of .org. He spent decades in ICANN policy work, beginning as director of policy at Network Solutions.

Hosted by Justin Beals, founder and CEO of StrikeGraph.

#CybersecurityAwarenessMonth #InternetGovernance #AIGovernance #AISecurity #SecureTalk

CMMC wasn't suspended. Only one piece of it was, and the defense contractors acting like the whole program went away are the ones taking on the most risk.

Eighty days after the DoD CIO paused CMMC Phase 2, the 60-day review has closed, a class deviation has written the pause into contracts, and the Reform Task Force's recommendations are still not public. Contracting officers and primes aren't waiting. Some solicitations now require a posted SPRS score before you can even open the RFP.

In this SecureTalk panel, host Justin Beals sits down with three NIST SP 800-171 practitioners (a C3PAO strategist, a lead CCA and instructor, and a former Navy cryptologist turned enclave provider) to separate what DoD actually decided from what the internet decided for it.

WHO THIS IS FOR

Defense contractors and subcontractors, compliance and IT leaders preparing Level 1 or Level 2 self-assessments, MSPs and MSSPs serving the DIB, and anyone trying to make sense of CMMC after the Phase 2 pause.

TIMESTAMPS

00:00 Why this episode is a panel

00:39 Meet the guests

04:11 "We're NIST 800-171 experts, not CMMC experts"

05:16 80 days after the pause: where things stand

07:21 What hasn't changed: Level 1 and Level 2 self-assessments

08:26 Myth: "CMMC was suspended"

11:33 No SPRS score, no RFP documents

14:08 CMMC is a floor, not a ceiling

14:26 The FAR overhaul and retired clauses still in contracts

18:29 Myth: "CMMC got pushed to November 2028"

22:02 The class deviation most people misread

24:14 What a defensible self-assessment looks like

29:06 Reddit, Discord and the misinformation problem

33:30 Myth: "My MSSP handles CMMC for me"

36:40 Myth: "Our policies prove we're compliant"

40:32 Myth: "We bought an enclave, so we're done"

44:07 Where the real risk lies now

51:13 How to choose outside help

SOURCES REFERENCED

- DoD CIO memo suspending CMMC Phase 2 (July 13, 2026), via Federal News Network: https://federalnewsnetwork.com/cybersecurity/2026/07/pentagon-suspends-cmmc-phase-two-requirements-launches-review-of-program/

- Sept. 3 acquisition memo codifying the pause, via MeriTalk: https://www.meritalk.com/articles/dod-codifies-pause-of-cmmc-phase-2-dod-cio-says-more-work-needed-on-cmmc/

- 32 CFR Part 170, the CMMC Program rule (phased rollout in §170.3): https://www.ecfr.gov/current/title-32/subtitle-A/chapter-I/subchapter-G/part-170

- NIST SP 800-171 Rev. 2: https://csrc.nist.gov/pubs/sp/800/171/r2/upd1/final

- NIST SP 800-171A, the assessment methodology: https://csrc.nist.gov/pubs/sp/800/171/a/final

- CMMC Program Final Rule: https://www.federalregister.gov/d/2024-22905/p-1876

- Free executive level class: https://www.evolvedcyberacademy.com/courses/Rev3Leadership

- CMMC Level 2 Self- Assessment: Step-by-step with Templates: https://www.strikegraph.com/blog/conduct-cmmc-level-2-self-assessment

WHAT YOU'LL LEARN

The requirement never moved. The pause hit third-party C3PAO assessments, not NIST SP 800-171 compliance and not Level 1 or Level 2 self-assessments. Brian Hubbard sets the bar at a self-assessment you could defend "if the DIBCAC walked in the next day after you posted your score."

November 2028 is not a new deadline. Logan Therrien explains it has always been the end of CMMC's four-year phased rollout. Reading it as a two-year reprieve is one of the costliest misreadings in the market.

The market is moving faster than the policy. Vince Scott compares it to oil moving through the Strait of Hormuz: there are months of supply in the system before anyone feels the change. Primes are already asking to see SSPs and SPRS scores.

GUESTS

Logan Therrien, Chief Strategy Officer, Kieri Solutions (C3PAO); 24-year military veteran

Brian Hubbard, President, Evolve Cyber; Lead CCA and CMMC instructor; 40+ years in cybersecurity

Vince Scott, CEO, Defense Cybersecurity Group; retired Navy cryptologist

HOST

Justin Beals, founder of Strike Graph and host of SecureTalk, a podcast focused on security news, innovation, and excellence.

🔔 Subscribe for conversations where cybersecurity, compliance and national security meet.

#CMMC #NIST800171 #DefenseIndustrialBase

She helps write the rules the entire U.S. defense industrial base gets assessed against — and she's telling you compliance is the floor, not the finish line.

Victoria Yan Pillitteri leads the Risk Management Framework/FISMA team at NIST and co-chairs the Joint Task Force uniting DoD, the Intelligence Community, and civilian agencies on one cybersecurity framework. In this episode, she and Justin Beals go inside how NIST actually builds SP 800-53 and 800-171 — what gets cut, what stays, and why "just copy the control language" is a losing strategy for anyone trying to pass an assessment.

In this episode:
Why 853 is "the Cheesecake Factory menu" of cybersecurity controls — and why that's a feature, not a bug
The real difference between NIST 800-171 Rev 2 and Rev 3, and why "organization-defined parameters" changed everything
Why writing your own control (not just quoting NIST's language) is the only way to actually pass an assessment
How FedRAMP 20x, OSCAL, and continuous monitoring are quietly replacing the point-in-time ATO
NIST's upcoming AI control overlays for predictive, generative, and agentic AI systems

Chapters
00:00 Introduction
00:34 The purpose of NIST standards and measurement science
02:24 Cybersecurity outcomes as a Rosetta Stone
03:14 The challenge of measuring risk in cybersecurity
04:55 Frameworks as operating systems for risk management
06:58 The iterative process of developing cybersecurity standards
08:11 Interpreting control statements for organizations
09:36 The importance of tailoring controls to risk profiles
12:30 The relationship between compliance and good risk management
14:37 The development process of cybersecurity standards
17:27 Differences between Rev2 and Rev3 of NIST 800-171
20:01 Broad versus specific requirements in cybersecurity controls
22:36 Supporting small businesses with guidance and tools
27:23 The balance between prescriptive and flexible standards
30:24 Cybersecurity in public-private partnerships
34:53 Moving from point-in-time to continuous authorization
40:23 AI risks and the development of tailored controls
44:53 The future of cybersecurity standards and AI security

Resources referenced:
NIST SP 800-53 (Security and Privacy Controls) — [link]
NIST SP 800-171 Rev 2 & Rev 3 (Protecting CUI) — [link]
NIST Risk Management Framework — [link]
NIST Cybersecurity Framework — [link]
NIST AI Risk Management Framework — [link]
FedRAMP 20x Program — [link]
OSCAL (Open Security Controls Assessment Language) — [link]

#NIST80053 #NIST800171 #CMMC #FedRAMP #CyberCompliance #RiskManagement #CUI #SecureTalk

An agent anyone could talk to just pulled the levers on one almost nobody could reach — and it happened inside the crown jewels: a live code repository.

Google gave its AI agent human-level trust — and paid for it.
When Dan Lisichkin, a researcher at Pillar Security, started mapping every Google repository running an embedded coding agent, he wasn't hunting for prompt injection — he was hunting classic CI/CD bugs. The AI angle showed up almost by accident, flagged by his own automation. What he found inside Google's Agent Development Kit repository was a low-privilege issue-triaging bot commenting on GitHub *as a trusted collaborator* — a status that should be reserved for humans the maintainers know. As Dan puts it, describing the moment his manager pushed back on downplaying the find: *"this is an agent triggering another agent... this is like no one talked about this before."*

The prompt injection wasn't the hard part — weaponizing it was.
Dan walks through Pillar's CFS framework (Context, Format awareness, instruction Salience) and how he literally used Google's own CONTRIBUTING.md file as the blueprint for the injection that would slip past the triage agent undetected. From there, one gated comment — normally reserved for trusted maintainers — was enough to trigger a second, far more privileged agent.

This isn't a bug you patch once — it's a new attack surface.
Dan's read is blunt: multi-agent systems create "weird machine" behavior — undefined states nobody designed for, not flaws in a specific line of code. He and Justin dig into why bolting more rules onto a non-deterministic system is Sisyphean, why bot identities need database-row-level granularity instead of human-style trust, and why Dan — a former malware researcher — thinks mandatory human-in-the-loop is often the wrong answer at scale.

Chapters: 

00:00: Cold Open: The Agent That Wasn't Supposed to Talk**
- Google's public triage bot and the collaborator-status anomaly
- Why "an agent triggering another agent" had never been formally described before

04:12:  Building the Hunt: Automation Over Manual Bug-Hunting**
- Dan's CI/CD vulnerability scanner, built on top of Claude Code
- How an AI-generated "AI agent injection" tag became the whole story
- Reference: [Simon Willison — "The Lethal Trifecta for AI Agents"]

14:30: The Exploit: Contribution Guidelines as an Attack Roadmap**
- Google ADK repository, the PR-triaging agent, and the CONTRIBUTING.md file used as a weapon
- Pillar Security's CFS framework for indirect prompt injection (Context, Format awareness, Salience) — [Pillar Security Blog: Autonomy of Indirect Prompt Injection]
- Why jailbreaking ≠ what Dan is doing — "I'm not trying to break the wall, I'm trying to walk through the door it left open"

28:05: Impact: Two Bugs, Two Verdicts**
- GitHub token exfiltration, PR/issue metadata manipulation, and the fake "looks good to merge" trail
- The second bug: a GCP service account and code-execution potential — "there was more juice on that one"
- Why Google didn't pay a bounty — and why that answer is more interesting than the bug itself

38:50:  Identity, Granularity, and the Human-in-the-Loop Debate**
- Why bot identities need GitHub App/Actions scoping, not personal-access-token trust
- The case *against* blanket human-in-the-loop — review fatigue, OpenClaw, and "people are going to do this anyway"
- SolarWinds, CryptoLocker, and why Dan thinks this is a closed-gap problem, not an open one

Resources: 
Lisichkin, D. (2026, August 3). I'll just call you: Agent-to-agent privilege boundary failures in CI/CD on Google's ADK repository. Pillar Security. https://www.pillar.security/blog/ill-just-call-you-agent-to-agent-privilege-boundary-failures-in-ci-cd-on-googles-adk-repository

https://danusminimus.github.io/

#aiagents #security #promptinjection #google #defcon #vulnerability
---

Communities are blocking $130 billion in AI data centers while the entry-level jobs that used to train the next generation of engineers quietly disappear.

Chapters: 

00:00:  The Backlash: 800 Groups, 49 States, $130B Blocked
Gallup: 71% of Americans don't want a data center built near them (Gallup)
Data center opposition tracker, Q1 2026 filings (referenced industry opposition data)
CMMC as precedent for regulating critical infrastructure (DoW CMMC Phase 2 program)

04:30:  Why AI Is "Eating Our Young" in Education**
Fran Berman & co-author's unpublished piece on the fraying mid-career pipeline
Better Tech (MIT Press) — Chapter appendix: AI classroom syllabus and exercises

14:00:  Tech as Critical Infrastructure, Not a Religion
Better Tech prologue: treating tech like food, water, roads, and the power grid
GDPR (EU, 2018) as a case study in regulation done right — and its limits
Vermont's data broker law as a case study in weak enforcement

26:00: Design Can't Be Bolted On Later**
Self-driving cars and the hidden environmental cost of full autonomy
Attack surface risk: denial-of-service on connected, self-driving fleets

34:00: Governing the Hybrid Human-AI Society**
EU AI Act — risk-tiered regulation: unacceptable, high-risk, low-risk categories
U.S. Equal Employment Opportunity Commission guidance on algorithmic hiring

41:00: The Hype Curve and the Data Center Reckoning**
Western Massachusetts communities rejecting new AI data centers
Efficiency vs. quality of life — Berman's closing argument

Communities are saying no to AI's biggest infrastructure bet.In the first quarter of 2026 alone, local opposition blocked or delayed 75 data center projects worth roughly $130 billion — nearly matching all of 2025's total in a single quarter. Dr. Fran Berman, former head of the San Diego Supercomputer Center, argues the fix isn't more hype, it's precedent we already have. She points to CMMC itself: "If we can look at the defense supply chain and say this is critical infrastructure, it has to meet a bar, then we can look at the trillion dollars of compute being built into the middle of American life and say the same thing."

AI isn't just displacing jobs.  It's starving the pipeline that builds senior engineers. Berman's sharpest warning is about who trains the next generation of professionals when entry-level coding and writing jobs — the ones junior people used to cut their teeth on — get automated away. She compares it to a surgeon who's never had supervised time in the operating room: "Unless you have that experience and the mentorship of more senior professionals, it's really hard" to develop the judgment senior engineers rely on.

Good regulation needs more than a law on the books. Drawing on her book Better Tech (MIT Press), Berman walks through why GDPR worked where Vermont's data broker law didn't — and previews how the EU AI Act's risk-tiered approach (unacceptable, high-risk, low-risk) could become the model for governing hybrid human-AI decision-making, where, as she puts it, "the only accountable entities are humans."

✍️ About the Author
Dr. Fran Berman is an award winning-data scientist, pioneer in public interest
technology, and community leader and builder. She directs the Public Interest
Technology Initiative at UMass Amherst and is a Faculty Associate at the Berkman
Klein Center for Internet and Society at Harvard. Berman is former head the San
Diego Supercomputer Center and served as Vice President for Research at
Rensselaer Polytechnic Institute. She currently serves as a Trustee of the Alfred
P. Sloan Foundation and is a popular regular panelist on public radio’s WAMC
Roundtable with 400,000 monthly listeners in seven states. For more information,
see https://www.franberman.com.

Link to the book: https://mitpress.mit.edu/978026205488...

Which AI agents can access what? Are those permissions even right? And can you stop a compromised agent mid-action? Okta's Dan Cinnamon says most enterprises can't answer any of the three.

Dan Cinnamon has built software, run SAP GRC without an implementation partner, and now architects identity for one of the industry's biggest players. In this conversation with Justin Beals, he lays out why "discoverability" — simply knowing what agents exist and what they're touching — is the single biggest blind spot in enterprise AI right now, and why there's no silver bullet coming to fix it. They also dig into passkeys as a rare win-win security standard, the maturing MCP spec, and where the hard line on agent containment should actually sit.

**Timestamps:**

0:00 Intro

1:20 From writing code to securing identity

4:45 Passkeys: the security/usability unicorn

8:30 The SAP GRC re-implementation story

14:10 Attribution and the agentic AI identity gap

18:55 How fast MCP has matured—and what's next

23:40 The 3 unanswered questions every enterprise faces

27:15 Granular identity vs. inherited permissions

32:00 Containment: moving controls closer to the data

36:45 Consent, provenance, and healthcare AI

40:30 Closing thoughts

#CISO #AIstrategy, #enterprise #AIsecurity, #agentic #AIgovernance, #Okta #MCPstandard,  #zerotrust #AI agents

The Pentagon paused CMMC Phase 2 with zero warning — and half the defense industrial base is celebrating for the wrong reason.

When the Department of War suspended CMMC Phase 2 rollout with no notice, panic spread fast across the defense contractor community — but the requirement to secure CUI never went away. In this special roundtable, host Justin Beals brings together three CMMC insiders — Logan Therrien (C3PAO Chief Strategy Officer, retired Navy submariner), Lance Arnold (30-year industry veteran, just completed his own CMMC Level 2 journey), and Brian Hubbard (President, Evolved Cyber Solutions, CMMC assessor since 2015) — to separate what actually changed from what didn't.

They break down the difference between the assessment requirement (paused) and the security implementation requirement (still very much alive under NIST 800-171), why "self-assessment" doesn't mean "no requirement," and what small businesses and primes should do right now instead of waiting for clarity that may not come for months.

Sources Referenced: 
DFARS 252.204-7021 (CMMC assessment clause)

DFARS 252.204-7012 (NIST 800-171 compliance clause)

DFARS 252.204-7019 (SPRS scoring requirement)

32 CFR Part 170 (CMMC Program Rule)

32 CFR Part 48

NIST SP 800-171 / NIST SP 800-172

Half the room at Chris Cochran's leadership dinners still calls themselves AI skeptics. He argues they can't afford to be — because the adversary already isn't.

Chapters: 

00:00 — Intro

02:49 — NSA/threat intel → AI security, storytelling

09:55 — Why leaders feel stuck / no roadmap

14:41 — Three pillars (Protect/Utilize/Govern)

17:00 — Governance as the real foundation / shadow AI

21:37 — Evidence-based scoring vs. pass/fail

26:27 — EU AI Act

28:44 — Fable/Mythos, Project Glasswing access controls

33:18 — Token subsidies, self-hosted models

37:52 — Data poisoning & context poisoning

40:12 — Iron Man suit framing

42:38 — Close: what's next

Most companies chase certifications to win deals — but what actually keeps customers is something no audit can measure.

In this episode, vCISO David Grazer makes the case that trust is a measurable economic asset hiding in plain sight: your customer retention rate. Drawing on 15+ years inside high-growth tech companies, David explains why compliance frameworks are customer acquisition tools, not retention strategies — and how the gap between the two is costing businesses more than they realize.

This episode is for founders, security leaders, and C-suite executives who want to connect their security and privacy programs to real business outcomes.

You'll learn:

→ Why a SOC 2 or ISO 27001 certification is only the beginning of earning customer trust

→ How customer churn functions as one of the most honest security metrics available

→ Why MFA and common security controls often fail the users who need them most

→ What "Trust by Design" looks like in product development and AI programs

→ How to translate security risk into language that resonates with your CFO

Chapters

00:00 Introduction to Secure Talk and Trust

03:42 David Grazer's Journey into Security and Privacy

08:09 Navigating Compliance and Customer Trust

12:49 The Role of Consulting in Security

18:07 Trust as a Measurable Economic Asset

23:42 Identity Management in the Entertainment Industry

26:09 The VC SO Model and Its Impact

29:13 The Evolution of Compliance Conversations

33:17 Exploring the Intersection of Technology and Society

🔔 Subscribe to SecureTalk for weekly conversations at the intersection of cybersecurity, compliance, and business strategy.

#cybersecurity #compliance #CISO #trustbydesign #vciso #informationsecurity #GRC #dataprivacy