AI Security Startup PR: How to Pitch a Company That Sits Between Cybersecurity and AI to Both Beats Simultaneously

If you're building a company that secures AI systems, uses AI agents to run security operations, or detects AI-generated threats, you're not just a security startup. You're not just an AI startup either. You're both, and that creates a PR problem most operators aren't equipped for.

The security beat and the AI beat are two distinct editorial cultures with different story criteria, different analyst maps, and different things that make a reporter pick up the phone. When you send the same pitch to Dark Reading and TechCrunch AI, one of two things happens: the security journalist thinks your pitch sounds like product marketing with no threat data, or the AI journalist thinks you're too deep in the weeds of CVE analysis to be relevant to their enterprise reader. Both ignore you.

Running PR for an AI security company requires two genuinely different pitches, built from the same underlying story, operating on the same timeline. That's what this post covers.

Why 2026 Is the Year This Problem Gets Expensive

AI security isn't a niche anymore. At Black Hat USA 2026, 35 of 121 briefings focused on AI security, AI red teaming, or LLM-assisted offensive techniques — 29% of every talk at the world's largest enterprise security conference. An exhibitor analysis found that 235 of 450 vendors on the show floor marketed AI or agentic capabilities. Seven of the ten RSAC 2026 Innovation Sandbox finalists either secured AI systems, used AI as a core security engine, or addressed attacks amplified by AI.

This means journalists on both the AI beat and the security beat are swimming in pitches from companies claiming to be at this intersection. The category is noisy. The journalists covering it have seen every variation of "we use AI to stop AI threats" from 200 startups this quarter. Getting through requires understanding that these two groups of reporters are not the same person with different email addresses. They're different audiences with different professional incentives, different editorial constraints, and different definitions of "a good story."

What the Security Beat Actually Wants

Security journalists at Dark Reading, The Record, CyberScoop, SecurityWeek, and SC Media are serving readers who are CTOs, security architects, and CISOs. The editorial culture is practitioner-first. These reporters are interested in novel applications, technical breakthroughs, significant research findings, and real proof points. They connect that interest to very specific types of evidence.

The highest-value asset you can hand a security journalist is original threat research. A report on a previously undocumented attack technique, a newly tracked threat actor, or a dataset that quantifies a risk most defenders aren't measuring. Vendors who conduct and publish original research get contacted by journalists. Vendors who only comment on other people's research wait in line.

What doesn't work: a pitch that says your product uses AI to detect threats faster, without a single MITRE ATT&CK mapping, no IOC list, no CISO quote from a production deployment. Security journalists see through vendor positioning immediately, and frankly they've been burned by it enough times that a pitch without hard proof gets deleted before they finish reading. A Dark Reading reporter wants the MITRE ATT&CK mapping and the IOC list. A Forbes security desk reporter wants a business-impact number and a company name. Those are different pitches even within the security beat.

Useful proof point types for security-beat pitches:

• Original threat research with quantified scope
• CVE commentary tied to a vulnerability class your product addresses
• Coordinated disclosure narratives where you discovered the vulnerability
• CISO-attributed case studies with specifics (threat reduced, time saved, incident avoided)
• Red team or pen test findings from production environments
• Incident response datasets your team generated

One thing I've been saying directly to founders I work with: if your security PR program can't point to at least one original research asset per quarter, you don't have a security PR program. You have a press release function. Those are different things.

What the AI Beat Actually Wants

AI reporters at TechCrunch, MIT Technology Review, The Information, Decrypt's AI desk, and Wired are operating with a completely different editorial lens. Their readers want to understand what AI can and can't do, who is winning the race to deploy it at scale, and what the real-world consequences look like at the enterprise level.

For an AI security company, the angle that lands on the AI beat is usually about AI system behavior, not about the attackers targeting it. It's about how AI agents behave in production, what goes wrong when they're misconfigured, and what governance looks like when you have hundreds of AI agents running across HR, CRM, file storage, and communication platforms simultaneously. The Microsoft Black Hat keynote framed this well: when an AI agent is trusted across your entire infrastructure stack, compromising that agent isn't one breach.

AI reporters care about:

• Capability benchmarks with honest comparisons (don't oversell)
• Enterprise deployment realities, not proof-of-concept demos
• The governance and compliance story behind autonomous AI systems
• Policy implications when AI makes consequential security decisions without human review
• Research that reveals how AI systems fail, not just how they succeed

MIT Technology Review has been explicitly bolstering its AI and security coverage, and its editorial guidelines are clear that their reporters need scoops, deep enterprise reporting, and sharp analysis. Commentary-only pitches go nowhere. You need a real finding or a real story to offer. The Information is even more demanding: they're looking for market intelligence that their subscriber base, mostly institutional investors and senior operators, can act on.

The gotcha nobody warns you about: AI reporters at most tier-1 outlets will fact-check your capability claims against public benchmarks and academic papers. If you claim your model performs 40% better than the baseline, they will ask which baseline, against what dataset, with what methodology. If you can't answer that, you've damaged the relationship and you're not getting a second chance.

Building the Dual-Track Pitch Program

The mechanics of running these two tracks simultaneously without the message falling apart come down to a single discipline: shared narrative, divergent proof points.

The shared narrative is the one true thing your company believes about the state of AI security that competitors aren't saying. It's not a tagline. It's a specific claim about the world that your data or your product existence proves. Geordie AI at RSAC 2026 built their narrative around "air traffic control for AI agents." That claim works in a Dark Reading context (here's how agents get exploited without visibility) and in a TechCrunch context (here's why enterprises need a governance layer before they can scale agentic AI). Same company. Two legitimate stories from the same core position.

Once the shared narrative is locked, you build out divergent proof points for each beat:

Security track proof package: - Threat research report with original data - CISO reference who will speak on the record about a specific incident or risk reduction - Technical brief with MITRE ATT&CK or equivalent framework mapping - Incident timeline or case study with enough specificity that the reporter can independently verify the attack pattern

AI track proof package: - Benchmark data with clear methodology disclosure - Enterprise deployment case study with scale numbers (number of agents monitored, volume of decisions made, false positive rate) - Policy or governance angle tying your product to regulatory developments (EU AI Act, NIST AI RMF, SEC disclosure rules) - A researcher or founder who can articulate a systemic point of view, not just a product pitch

The pitch you send to Dark Reading mentions the CISO proof point in paragraph one. The pitch you send to TechCrunch AI mentions the deployment scale and governance angle in paragraph one. The reporter at each outlet gets a pitch that reads like it was written for them, because it was.

The Embargo and Timing Problem

Running two simultaneous PR tracks means you're managing embargo coordination with reporters who don't know each other, have different publication cycles, and are competing against each other for some of the same audience. A security journalist at The Record and an AI reporter at TechCrunch are both publishing to enterprise technology buyers. If your embargo drops to both at the same time and both stories run on the same day, you've actually helped both outlets. The coverage confirms the story from two angles.

Where founders get into trouble is treating embargoes as the same thing across both beats. Security reporters are accustomed to standard 48-to-72-hour embargo windows for threat research, often structured as a two-page journalist brief with the one-sentence finding, affected scope, researcher credibility, and patch status. AI reporters are more variable. Some want an exclusive window. Some prefer a briefing call with no written material in advance. Some at The Information will only run proprietary market intelligence they've independently confirmed.

The practical move: brief both tracks during the same 72-to-96-hour embargo window, but brief them separately, with tailored versions of the material. Never send the security brief to the AI reporter and vice versa. A security-dense brief full of CVE references landing in the inbox of an AI journalist who covers enterprise deployments reads as the wrong company for their beat.

Common Pitfalls That Kill Dual-Track Programs

Leading with the product on both sides. Security journalists don't write about products. They write about threats and defenses. AI journalists don't write about products. They write about capabilities, deployments, and consequences. If your pitch leads with "our platform uses AI to provide real-time threat detection," both reporters will stop reading. Lead with the finding or the story, not the product.

Using AI-generated commentary pitches. I'll be direct about this one: a cybersecurity reporter at Cybersecurity Dive said publicly that AI-generated or AI-assisted commentary is immediately visible, and it "harms you, it harms us, it harms our readers." The security beat skews technical and experienced. These reporters have read enough vendor pitches to recognize templated AI output on sight. Sending it doesn't just get your pitch ignored. It tags you as someone not worth engaging with next time.

Conflating the analyst audience. Gartner, Forrester, and IDC cover cybersecurity with dedicated practices. They want to understand where you fit in the security ecosystem, what you integrate with, and what use cases you address. They're not the same audience as Kuppinger Cole or the AI-focused analysts at IDC's AI practice. Running analyst relations for an AI security company requires mapping which firm covers your category, not just which firms are "big."

Treating APAC as a single market. If your company has any presence or ambition in Asia, both beats fracture further. Japanese security media cares about METI compliance angles. Korean security coverage skews toward critical infrastructure. Indian enterprise media wants the business transformation story, not the threat research deep-dive. The dual-track problem becomes a six-track problem, and that's before you account for the fact that some Asian security reporters cover cybersecurity from a policy angle, not a practitioner angle.

Calling it a "game-changer" anywhere in the pitch. Both communities have high tolerance for technical detail and low tolerance for marketing language. If that word appears in your pitch, the reporter has already decided you don't understand their beat.

What the Best-Positioned AI Security Companies Do

The companies that earn coverage on both tracks in the same quarter share one operational practice: they produce original research on a cadence. Not a product announcement dressed up as research. Actual new findings their team generated from production data or controlled testing. That research gives security reporters something to report, gives AI reporters a dataset to interrogate, and gives you legitimate authority on both sides simultaneously.

The GAIA case from my own client work is instructive here, even though GAIA operates in decentralized AI rather than enterprise security. We drove placement in Forbes, Decrypt, and Benzinga from the same founder story by building out divergent proof points for each outlet. Forbes wanted the vision of what decentralized AI agents could do for the broader internet. Decrypt wanted the technical architecture and the node count as validation. Neither pitch shared more than the narrative spine. That's the discipline.

For an AI security company, the equivalent looks like this: your original threat research lands in Dark Reading and SecurityWeek. Your enterprise deployment story with CISO attribution lands in TechCrunch and The Information. Your founder's systemic point of view on AI governance lands in MIT Technology Review or Wired as a bylined piece. Three placements, three different story angles, one coherent underlying company position.

builders build, PR's pitch lol Shilika Jain on X, 28 Mar 2025

That's flippant but it's also exact. Your job as a founder is to build the company and generate the evidence: the threat data, the CISO relationships, the deployment metrics. The PR job is to translate that evidence into story angles that fit two editorial cultures simultaneously, without making the company sound like a different company to each audience.

That translation is not trivial, and it's not something a generalist agency that also handles SaaS and consumer apps will execute cleanly. The security beat requires pre-existing reporter relationships and a demonstrated understanding of how threat research moves through that community. The AI beat requires knowing which reporters are covering enterprise deployment versus model research versus policy, and pitching accordingly. Running both from the same desk, without sacrificing quality on either side, is exactly the kind of work where embedded fractional support outperforms a junior account team at a large agency.

If you're six to ten weeks away from a funding announcement, a product launch, or a major research drop, and you haven't yet mapped which reporters on both beats will receive your pitch and why, that's the gap worth closing first.

---

Book a 30-min teardown with Shilika: https://calendly.com/shilikajain/30min/