Fritz Abelard

Teams adopting AI coding assistants are trying multiple tools at once to speed up delivery, creating attack paths that legacy security tools were not designed to catch. The response they describe is what they call secure at inception: protection embedded in the coding loop, making security invisible in the way a customer said it should be—present, but not something developers have to stop and operate.

In the post on the LiteLLM supply-chain incident, Fritz Abelard recalls a customer saying "security should be invisible" and explains that the point is to embed security, not eliminate it.

• They read the incident—in which malware entered through a dependency and harvested credentials across systems it touched—as an example of what happens when security is not embedded.
• An earlier post on Snyk's Secure At Inception turns the principle into three concrete claims: real-time scanning at the point of code generation, visibility into AI components inside enterprise software, including generative AI and MCP, and an experimental scanner for MCP-specific vulnerabilities.

Each statement of the general principle arrives alongside a Snyk product launch, so read the security reasoning separately from the vendor claim in the same post.

Read original · Original 2

AI excels at boilerplate, synthesis and brainstorming; human expertise supplies final verification and domain-specific context. Model output can also fall back on oversimplified patterns that sound plausible, making a deliberate verification loop more useful than blanket trust or distrust.

The post asks whether developers would blindly commit code from GitHub Copilot or Cursor, then lists four habits: treat AI output as a draft; learn each tool's strengths and weaknesses; keep a human in the loop for the final check; and stay skeptical enough to catch and correct patterned errors. It closes by asking readers how they build that verification loop into their workflows.

Read original

They argue that agentic systems, where models call tools and act in the development loop, open attack surfaces that existing static and runtime analysis do not cover. The posts introduce each threat alongside the response, giving readers both the risk and the check.

The clearest case is Toxic Flows, a class of vulnerabilities that, according to Abelard, makes agentic AI systems susceptible to prompt injection and Model Context Protocol exploits, based on research from Invariant Labs.

• The post pairs the finding with Snyk's Toxic Flow Analysis framework and the MCP-scan tool.
• The RSAC post adds a live example—the LiteLLM dependency compromise—and maps it to checks across agentic development and agentic applications: the tools agents connect to, the code they produce, their behavior in the development loop and validation of AI-powered apps before release.

Read original · Original 2 · Original 3 · Original 4

In a point-solutions business, the role stays shallow and expertise does not compound; on a platform, the work accumulates. They frame this as a career decision with a consequence for companies, too: firms that treat technical success managers as replaceable parts are losing people who are already being recruited elsewhere.

The talent post opens by saying talented technical success managers are being secretly hunted while most companies still treat them like replaceable parts.

• Abelard presents a view they describe as coming from three seats at the table.
• The RSAC post applies the idea directly: be the technical success manager who builds compound expertise at companies where the roadmap matches the actual problem.
• They contrast that with a point-solutions mindset that cannot keep up when the threat surface is moving quickly.

Read original · Original 2

Abelard’s method: incident, then principle, then product fit.

Abelard describes spending RSAC week at their desk while following the LiteLLM supply-chain incident, in which malware entered through a dependency and harvested credentials across the systems it touched.

• They connect the incident to two earlier themes: “platform gravity” in the technical success manager talent war, and a customer’s line that “security should be invisible,” meaning embedded rather than bolted on.
• They then cover Snyk’s RSAC announcements: Agent Scan, Snyk Studio, and Agent Guard for agentic development; and Agent Red Teaming plus Snyk API & Web for agentic applications.

For brands planning a campaign

Useful for campaigns that need a practitioner to connect a current incident to a vendor roadmap, especially in supply-chain or AI-agent security.

This short, reusable framework applies to any AI-assisted workflow.

Abelard argues that AI is a powerful collaborator, but not an infallible authority, whether used for content, productivity, or code generation.

• The post makes four points: verify AI output and treat it as a draft; understand each tool’s strengths and weaknesses; use AI to augment rather than replace people, with a human-in-the-loop model; and scrutinize and correct AI’s tendency to fall back on oversimplified, pre-existing response patterns.
• They close by asking readers how they establish that verification loop.

For brands planning a campaign

For AI productivity or developer-experience campaigns, this offers a balanced, verification-first angle for a practical guide.

A useful introduction to MCP, prompt injection, and toxic flows.

Abelard writes that AI adoption is exposing new attack surfaces and describes “Toxic Flows” from Invariant Labs research—a class of vulnerabilities that make agentic AI systems susceptible to prompt injection and Model Context Protocol (MCP) exploits. They announce Snyk’s Toxic Flow Analysis framework and the MCP-scan tool as a first-of-its-kind approach to detecting and mitigating these risks, and point readers to a blog covering both releases.

For brands planning a campaign

Useful for campaigns aimed at security teams evaluating AI agents: it names a specific threat model rather than discussing AI risk generically.

This shows how customer signals become product positioning, a pattern running through most of their posts.

Abelard reports from customer meetings that securing AI is becoming a top priority.

• Teams are “trying all the things” to improve developer experience and speed, while the era of vibe coding introduces new attack vectors they label toxic flow.
• They argue that legacy security tools are no longer sufficient and describe Snyk’s Secure At Inception as built on three innovations: deeply integrated, real-time scanning at the point of code generation; visibility into AI components in enterprise software, including generative AI and MCP; and an experimental scanner for AI-specific MCP vulnerabilities.

For brands planning a campaign

For developer-security campaigns, this is the clearest statement of the “security from inception” idea and can anchor a message about embedding protection in the coding loop.

This shows the professional identity behind the product posts and offers a starting point for understanding their audience and credibility.

Abelard writes that talented technical success managers are being secretly hunted while most companies still treat them as replaceable parts.

• They offer a perspective “from three seats at the table” on why companies that get this right will win.
• The post is shorter than the security-focused ones and includes a light personal aside about pups and P.F.
• Chang’s.

For brands planning a campaign

Relevant for employer-brand, customer-success community, or talent campaigns; less product-specific than the security posts.

Abelard names specific risks: prompt injection and Model Context Protocol exploits, a vulnerability class described as "toxic flows," and malware slipping in through a dependency in the LiteLLM supply chain incident.

• The recurring argument is that security must be present when code is generated, not added after release.
• The posts repeatedly tie that argument to product launches built around it.

Introduces each unfamiliar risk by naming the vulnerability class and the mitigation tool released against it, bringing the abstract threat and response together.

Several posts describe Snyk's own products. Treat the threat framing and the vendor claim in each post as separate evidence.

For brands planning a campaign

The strongest fit is a brief combining DevSecOps practice with modern AI security threats. A campaign could ask for a plain-language threat model of agentic systems, a walkthrough of how to assess risk in an AI coding assistant, or a worked example showing where a review checkpoint belongs in an AI-assisted build pipeline.

Examples: Source 1 · Source 2 · Source 3 · Source 4

The post addresses people using AI for content, productivity, and code generation.

A numbered list of four practices, each paired with a one-line reason, ending with a question inviting practitioners to describe their own verification loop.

For brands planning a campaign

Useful for audiences evaluating AI coding or writing tools who need a review process, not a verdict on the tools. A brief could ask for a verification checklist or a short internal policy draft for teams adopting AI assistants.

Examples: Source 1

Each post explains what was found, notes who is researching or monitoring it, and connects the findings to a broader argument about how quickly the threat surface is moving.

Summarizes the incident or research in a few sentences, credits the teams involved, then links to the primary coverage alongside vendor material.

For brands planning a campaign

Supports timely explainer content tied to a live security story, making it useful for a brand that can work on a short lead time. The value depends on speed rather than depth, so allow for a quick turnaround and don't assume the audience has read the underlying research.

Examples: Source 1 · Source 2 · Source 3

The argument is that the best technical success managers join companies whose products pull customers deeper into an ecosystem rather than solving a single point problem, and that companies treating these roles as replaceable parts are losing talent they do not realize is being recruited. Abelard writes from customer calls and describes drawing lessons from "three seats at the table."

First-person observations from day-to-day customer work, framed as patterns other people in the same role should act on.

These posts draw on personal experience; for a technical evaluation piece, ask for a more evidence-led tone.

For brands planning a campaign

The most personal and opinionated material in the corpus speaks to vendor-side audiences, such as customer success and security leadership, rather than buyers comparing tools. It suits campaigns about talent, career development, or partner enablement—not product evaluation briefs.

Examples: Source 1 · Source 2

The FedRAMP Moderate Authorization post is the clearest example: it frames the news as two buyer outcomes, enhanced security assurance and easier procurement and deployment for government entities. Posts on the AI Trust Platform and Secure at Inception follow the same shape, naming the speakers, components, and problem each component addresses.

Announcement, followed by a short list of components or benefits, then a link and hashtags.

The posts contain no adversarial or evaluative coverage of a competitor. For a comparison brief, ask for independent criteria and evidence from multiple products.

For brands planning a campaign

Well suited to launch amplification, partner announcements, or grassroots promotion when the brand already has a product story to carry. Less suited to independent comparison content: the posts align with the employer's messaging rather than testing it.

Examples: Source 1 · Source 2 · Source 3 · Source 4 · Source 5

It is the material's most specific single-topic post and the only one to address a regulated buying process directly.

Pairs the achievement with a checklist of what it means for customers, distinguishing security assurance from procurement speed.

For brands planning a campaign

Relevant for a brief aimed at government or regulated buyers, who care about authorization levels and procurement steps. A campaign could ask for a short explainer on what a moderate authorization means for a federal agency evaluating developer security tooling.

Examples: Source 1

Topics start with the problem, then introduce the product or practice. Several posts repeat a compact phrase such as "secure at inception" or "security should be invisible" to hold the argument together.

Consistent across their posts, making the format predictable to plan around.

For brands planning a campaign

A brand can brief this format directly: lead with a claim, add three to four supporting points and one concrete example, then link to a primary source. It suits LinkedIn-native campaigns and compressed explainers; the repeated phrase gives a campaign a reusable slogan hook.

Examples: Source 1 · Source 2 · Source 3 · Source 4 · Source 5 · Source 6

Their posts cover Snyk product announcements (Secure At Inception, Agent Security, Evo AI-SPM, FedRAMP Moderate authorization for Snyk for Government) and the company's monitoring of the LiteLLM supply-chain incident. Multiple posts carry the employer hashtag #LifeAtSnyk or #lifeatsnyk, and Abelard refers to earlier pieces they wrote that were published through Snyk channels.

Original disclosure

Posts carry the #LifeAtSnyk / #lifeatsnyk employer hashtag.

View example

The post credits the research to Invariant Labs.

View example

Both also argue that security value has to reach the customer, not just the research team.

Sorani's material reaches back to an offensive-security origin story and focuses on the economics of fixing vulnerabilities at scale, while Abelard focuses on the AI-era attack surface and the customer-facing technical role.

Read original · Original 2

Camacho works in embedded and automotive systems, where long product lifecycles, constrained hardware, and safety standards such as ISO 21434 set the constraints. Abelard works in AI-assisted software development, where coding assistants, agents, and model context shape the constraints.

Read original · Original 2

Both treat agentic activity as a present-tense threat, not a future one.

Manico is openly contrarian, arguing that traditional scanners and AppSec business models are collapsing, and cites incidents as evidence for a market thesis. Abelard writes from a vendor's customer-facing seat, pairing each risk with a specific product response and rarely taking a position against a competing approach.

Read original · Original 2

Both posts argue that detection and visibility are only step one, and that the hard problem is the "last mile" — fixing vulnerabilities at scale without slowing development — using telecom's last-mile investment as an analogy.

• For a CTO-facing explanation of where checks belong in an AI build pipeline, that is a closer match to what a training vendor sells: a workflow that gets findings closed, not another alert source.
• Sorani also describes their route into offensive security through studying and generating computer viruses, giving a brand a usable origin story for an explainer opening.

The two posts say nothing about AI coding assistants, prompt injection, Model Context Protocol, agentic tooling, or supply-chain attacks through dependencies.

• The AI threat surface central to Sorani's material is absent, so an AI-security campaign would need to bring its technical spine from the brand's own research and use Sorani for the remediation argument.
• Sorani's material is framed through co-founder conversations at their company, not standalone teaching posts, so the brand would need to shape the format.

Examples: Source 1 · Source 2

The posts cover cybersecurity across the full lifecycle of software-defined vehicles — telematics, ECU communications, over-the-air updates — and name the standard in play, ISO/SAE 21434.

• Camacho lays out the verification stack in concrete terms: static analysis early, dynamic analysis and fuzz testing during execution, penetration testing to validate defenses, secure coding, security gates in CI/CD, and AI-assisted verification with human oversight.
• For a CTO whose engineers ship firmware on long product lifecycles, that process-level answer meets their need more directly than commentary on AI coding tools.

The frame is automotive and embedded, so a general software or cloud security audience would have to translate it; the sales argument about engineering discipline and safety may not land with buyers who do not ship hardware.

• Neither post discusses AI coding assistants, agentic systems, or MCP-based exploits, so the AI-threat half of the campaign would need to come from elsewhere.
• The posts largely summarize the author's own published guides, so a campaign built on them risks repeating material the brand's readers have already seen.

Examples: Source 1 · Source 2

Their argument is that AI agents are collapsing the economics of both finding and fixing vulnerabilities, that standalone scanners, SAST, DAST, IAST and CVE tooling will increasingly be displaced by AI-native capability, and that standards such as OWASP ASVS and AISVS matter more, not less, because they define what has to be verified.

• Manico also takes on the "AI cyber is BS" position directly, citing an in-progress attack on RubyGems by automated package publishing.
• For a CTO deciding whether to reorganize security work around continuous verification, that is the stronger provocation, and the standards framing gives a training vendor curriculum hooks for a checklist.

Much of the argument is conjecture about where the industry goes next, and the register is adversarial, so a brand seeking a neutral educational tone would need to soften it and keep the claims attributed.

• The posts offer no practitioner how-to — no workflow, template, or checklist — so the brand would need to provide the campaign's downloadable asset.
• Manico's independence from vendor product messaging is a reason to choose them, but their material offers no product vocabulary if the campaign depends on it.

Examples: Source 1 · Source 2