# Chris H.

Chris H. writes and speaks about cybersecurity, with a focus on software supply-chain security and how AI changes security risks and defenses.

- Profile: https://b2binfluencer.marketing/creators/chris-h/
- LinkedIn: https://www.linkedin.com/in/resilientcyber
- LinkedIn followers: 82,194
- Median of last 8 posts: 14 reactions, 6 comments (checked 2026-11-02)
- Topics: artificial-intelligence, supply-chain, cybersecurity, ai-security, software-supply-chain-security, vulnerability-management
- Based on 8 dated posts, latest 2026-10-07

## Summary

Most useful for explaining AI safety and supply chain security research to security practitioners.

- Best for: A cybersecurity training company building a course on AI safety and supply chain security.
- Not for: Consumer privacy or family online safety brands wanting broad, non technical lifestyle content.
- Check first: Ask for audience survey results covering job role and seniority.

### Takeaways

- AI safety and cybersecurity are one problem, and both layers can fail.
- Human-in-the-loop oversight is often a rubber stamp.
- Finding vulnerabilities with AI has to be paired with preventing new ones.

## Main ideas

- **AI safety and cybersecurity are one problem, and both layers can fail**: Chris H. takes NVIDIA's "assume misalignment" framing and sets it beside the security maxim "assume breach." The argument is that perfect alignment of AI systems and agents is unlikely, that well-funded frontier labs are visibly struggling with it, and that this makes safety and security complementary rather than competing. The suggested approach is defense in depth: established security primitives plus emerging alignment practice, built on the assumption that either one alone will eventually fail.
- **Human-in-the-loop oversight is often a rubber stamp**: Chris H. argues that human-in-the-loop review frequently functions as a rubber stamp that makes teams feel safe without changing what happens, and that classifiers added as an extra guardrail in agent "auto modes" are not sufficient on their own. The evidence cited is Anthropic research that red-teamed Claude Code's Auto Mode and OpenAI's Guardian in Codex and bypassed the monitors using agent-generated prompt injection, multi-agent attacks and malicious compaction — a result the researchers themselves describe as unsolved.
- **Finding vulnerabilities with AI has to be paired with preventing new ones**: Chris H. works through Anthropic research that used an internal, unreleased model to find critical vulnerabilities in Chrome, Safari, Windows, Android and Linux, much of it with little or no human interaction and with multiple concurrent scanners, techniques and compute. The part Chris H. singles out is the closing argument of the research: locating flaws in existing code is only half the job, and stopping vulnerabilities from entering new code matters just as much. Chris H. calls this the new normal and frames prevention as the condition for gaining AI's development benefits without an explosion in attack surface.
- **Attackers are industrializing their operations, even when the tradecraft is old**: Chris H. describes a shift across 2026. In the first half of the year, in their reading, the story was AI industrializing vulnerability discovery while exploitation stayed fairly traditional. Over roughly the following eight weeks, they see attackers industrializing their own operations, adopting open-source agentic frameworks such as ARTEX and multi-agent architectures with limited or no human involvement, letting a single attacker run intrusions against hundreds of organizations at once, and in some cases using stolen inference. Chris H. keeps an important qualification: the techniques remain largely traditional, but the boost in time, skill and capability is real. CrowdStrike, Mandiant and Wiz are the sources cited for the attack reports.
- **Loss data, not scare narratives, should set vulnerability priorities**: In an interview with Jeremiah Grossman, CEO of Root Evidence, Chris H. tests the "vulnpocalypse" narrative against exploitation and loss data. Grossman's insurance claims data ties only about 600 CVEs to financial loss, and his defender-window research found attackers exploiting N-days in 2026 waited a median of 116 days after the patch, up from 24 days in 2018. The conversation contrasts Grossman's bottom-up, loss-first approach with a different zero-day tracking model, and covers warranties, software liability, the missing catastrophe database for cyber, and the question of where actuarial data for any security control can be found. These are Grossman's findings and positions as relayed by Chris H.; Chris H. closes by asking readers what it changes about their own backlog.
- **Identity is the thread running through SaaS and cloud, and agents lengthen it**: Chris H. describes earlier federal SaaS security work in which the hard part was telling personal accounts apart from enterprise identity provider accounts across hundreds of applications and thousands of users. The argument now is that SaaS platforms shipping AI and agents add more identities, tools and connected resources on top of users and apps that were already hard to inventory, so the identities to govern include people, service accounts and agents — and SaaS is where all three overlap first. Chris H. treats SaaS and cloud as one connected attack surface threaded by identity.
- **The supply chain now includes developer tool marketplaces**: Chris H. co-wrote a book on software supply chain security with Tony Turner, aimed at a foundational treatment of a complex topic, and notes its nomination for the Cybersecurity Canon while observing how quickly LLMs, agents, MCPs and skills have complicated the field. The practical version of the argument is that developer tools sit close to source code, cloud credentials and CI tokens, so a trusted extension marketplace is a supply chain target deserving the same scrutiny as npm and PyPI.

## Work to read

- [Co-authored book on software supply chain security](https://www.linkedin.com/posts/resilientcyber_software-transparency-supply-chain-security-activity-7513569619748605952-H6N6)
- [Assume Misalignment: why AI safety needs security thinking](https://www.linkedin.com/posts/resilientcyber_assume-misalignment-securing-ai-agents-beyond-activity-7513645722391535617-FmCJ)
- [Red-teaming agent auto modes](https://www.linkedin.com/posts/resilientcyber_red-teaming-auto-mode-weve-seen-the-rise-activity-7522269184974802944-X6BN)
- [An unreleased model finds critical vulnerabilities in widely used software](https://www.linkedin.com/posts/resilientcyber_critical-exploits-in-some-of-the-most-widely-activity-7522659139911831552-Maso)
- [Is the Vulnpocalypse real? An interview about exploitation and loss data](https://www.linkedin.com/posts/resilientcyber_is-the-vulnpocalypse-real-that-was-one-activity-7523111729460060160-XORH)
- [From AI-accelerated bug hunting to agentic attacks](https://www.linkedin.com/posts/resilientcyber_are-we-moving-from-ai-being-used-to-industrialize-activity-7523014222344245248-bO8z)
- [Trojanized Claude Code extension in the Cursor Marketplace](https://www.linkedin.com/posts/resilientcyber_a-trojanized-version-of-the-claude-code-extension-activity-7521887735662432256-c3kf)

## Brand relationships

- **Palo Alto Networks** (disclosed_partnership)
- **Wiley** (mention)
- **NVIDIA** (mention)

## Partnerships

Chris H. publishes as Resilient Cyber. Within the ten posts reviewed for this section, only one documents a vendor partnership: a post that names Palo Alto Networks as a Resilient Cyber partner and describes time spent with the vendor walking through the SaaS Security capabilities in Cortex Cloud 2.3. Elsewhere the material cites vendor research and interviews as sources rather than as partners, including CrowdStrike and Mandiant reporting on multi-agent attacks, Wiz on stolen inference, Anthropic on automated vulnerability discovery and red-teaming, and NVIDIA on AI alignment. No published sponsorship policy, rate card, format specification, or partnership inquiry route appears in the reviewed material, so no inquiry link is shown. A brand evaluating a paid placement has one disclosed example to work from and should confirm terms directly.

## Alternatives

- Demawu Apeti: Choose Demawu Apeti over Chris H. if the course needs a governance-first spine: integrating AI risk into ISO 27001, GRC, supplier assurance, change control, and business continuity processes rather than building a separate AI-governance track. The two posts show a governance and risk-management framing, with human oversight, accountability, continuous risk assessment, and auditability for industrial environments where AI can influence physical operations. The ACSP registration post adds a visible professional-standards signal. That fits a curriculum module on AI governance for regulated or industrial learners, where the buyer wants an assessor's process language more than attack teardowns.
- Tracy Levine: Choose Tracy Levine over Chris H. when the course needs third-party vendor risk or a legal-sector post-quantum module. The posts discuss cybersecurity leaders working with vendors to enforce rigorous security standards and provide assurances for shared data and systems, and SonKsuru aligning quantum-proof solutions for the legal profession, with Jack Rives welcomed to an advisory board. If the training buyer's learners are legal, compliance, or procurement-adjacent teams, or if the course needs a post-quantum transition angle for critical industries, Tracy Levine's material speaks more directly to those decisions than a curriculum built around agentic attack surface.
- Evan Powell: Choose Evan Powell over Chris H. when the course should center open-source AI security, detection engineering, and operating AI for security rather than supply chain history and governance. The posts describe Linux Foundation involvement, an Open Secure AI Alliance, DeepTempo's LogLM and Vigil open-source AI SOC, a case study with the NSA-established TAC for OT protection, and a current read on Jev as a System One model for triage, ATT&CK tagging, routing, and guardrails. If the buyer wants a builder's view of open-source tooling, transparency, and human-in-the-loop orchestration for AI detection, Evan Powell's two 2026 posts give that perspective with recent product and community details.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/