AI Is Widening the Security Perimeter—and Raising the Price of Trusted Defenders
Counterintelligence hiring at major AI firms and the spread of AI-assisted ransomware point to one shared pressure: defenders must now protect both scarce secrets and a far broader set of people-facing entry points.
By Theo Mercer · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess a human career history, credentials, or firsthand experience.
Key points
- Major AI firms are recruiting counterintelligence personnel as concerns about industrial espionage rise, while available specialist talent is described as scarce.
Sources: S1
- Ransomware researchers report increased claimed activity and say criminals are using large language models for tasks such as phishing and negotiation, even though successful groups remain largely human-operated.
Sources: S2
- The two developments expose different security problems: protecting concentrated high-value knowledge versus reducing scalable deception across an organization and its suppliers.
The security imbalance is moving in two directions
The evidence describes two distinct, connected pressures on organizations building or using advanced AI. Fortune reports that AI companies including OpenAI, Anthropic, and Nvidia have expanded counterintelligence hiring amid concern over industrial espionage and the treatment of AI as a national-security matter. Separately, WIRED reports that ransomware researchers are seeing criminals experiment with large language models in parts of the attack chain, particularly phishing and negotiation. These are not accounts of the same threat or the same response. One centers on access to strategic secrets and influential people; the other concerns the multiplication of credible, routine intrusion attempts against a much wider target pool.
A premium defense for a concentrated asset
The counterintelligence market described by Fortune is costly and selective. Citing Revelio Labs, the report says salaries for such roles at AI firms generally range from $200,000 to $500,000. It reports that Anthropic hired 33 employees for counterintelligence functions so far this year, compared with 14 last year and nine in 2024; OpenAI’s hiring rose to 22 in 2026 from nine in 2025; and Nvidia hired 16 people this year. A former Secret Service agent told Reuters that demand exceeds the supply of people with relevant experience. That points to a defense model dependent on a relatively small labor pool with government, investigative, and security experience.
Sources: S1
Sources: S1
The human route remains the hard part
Fortune’s reporting emphasizes that counterintelligence is not simply a technical-control function. A former FBI counterintelligence operative said approaches can begin through conferences, business introductions, drinks, or dating apps rather than a direct request for secrets. He identified fast-moving relationships, repeated questions about sensitive business, requests for introductions, and pressure toward secrecy as possible warning signs. The practical challenge is governance as much as detection: a security team needs a route to raise concerns when the person with the most sensitive access is also trusted, senior, or unwilling to accept intervention. Tools can monitor systems, but they do not by themselves resolve conflicts over executive judgment, privileged access, or organizational authority.
Sources: S1
Sources: S1
AI assistance is expanding the lower end of the attacker market
WIRED cautions against treating current ransomware as fully autonomous. Researchers cited in the report say fully AI-powered ransomware attacks have not emerged, and that successful ransomware groups are still largely human-operated. The reported change is more incremental but potentially broad: attackers appear to be using LLMs to create more effective phishing messages and assist negotiations, while researchers also describe uses or attempted uses in vulnerability exploitation, code checking, and maintaining alternative access to a victim. These tasks matter because social engineering often supplies initial access, and because language quality can make cross-border targeting more convincing.
Sources: S2
Sources: S2
Volume is a warning signal, not a settled measurement
The ransomware evidence comes with significant limits. WIRED says exact attack totals are difficult to establish because victims may not disclose incidents and attackers may exaggerate claims. Some researchers put alleged attacks posted on leak sites at nearly 3,000 over the summer and estimated an increase of 20 to 40 percent since the spring, but those are claims requiring de-duplication rather than a definitive victim count. Halcyon identified around 111 groups making attack claims between July and September, while examples of newer groups included one claiming 48 victims and another listing 71 victims in 20 days. The report explicitly notes that not all claimed attacks may be legitimate.
Sources: S2
Sources: S2
The connection is a dependency on people, not proof of one merged threat
Inference: the cross-source comparison suggests that AI may be stretching the security labor market at both ends. At the high end, firms with valuable models, chips, research, or strategic relationships are buying specialized judgment to defend against targeted human manipulation. At the broad end, organizations face a potentially larger flow of tailored phishing, extortion, and access attempts, including from smaller or less experienced operators. The shared dependency is human capability: employees who recognize deception, security teams able to triage it, and leaders willing to permit escalation. This is an inference from the reported hiring and attacker experimentation, not evidence that AI-assisted ransomware groups are conducting industrial espionage or that counterintelligence teams can substitute for ransomware defenses.
Openness and affordability become security design questions
This split matters for who can afford meaningful protection. A well-funded AI developer may recruit former intelligence and security officials, build dedicated investigative functions, and formalize executive-protection practices. Smaller companies, research groups, suppliers, and ordinary ransomware targets may not have comparable access to that expertise. Yet they can still be exposed through the same basic channels: persuasive messages, compromised remote-access tools, sensitive conversations, and relationships that cross organizational boundaries. An ecosystem that concentrates defensive knowledge inside a few large firms can leave dependencies elsewhere—contractors, customers, cloud administrators, and smaller technology partners—less prepared to identify or report suspicious activity.
What would change the assessment
The key question is whether AI experimentation becomes durable operational advantage rather than a burst of noisy claims. WIRED reports that some researchers expect inexperienced groups to burn out without the institutional knowledge and resources of established operations, and Recorded Future’s analyst said encryption-involving attacks may decline even as the total number seen this year exceeds 2025. Evidence of sustained, independently verified compromises by newer groups, or evidence that AI-enabled phishing materially improves intrusion success, would strengthen the case that the barrier to entry has shifted. On the defense side, more transparent evidence about how counterintelligence programs prevent compromise—and whether their practices can be adapted by organizations without elite hiring budgets—would show whether this premium response is broadly transferable.
Why it matters
Security strategy cannot treat AI risk as only a model-security or malware problem. The reported developments suggest a layered contest in which scarce specialist defenders protect high-value secrets while more organizations must withstand scalable, human-centered deception. The practical test is whether useful defensive capacity can spread beyond the best-resourced firms without relying on the same scarce experts attackers and strategic rivals are already forcing them to hire.