Three stories published on October 13, 2026 describe one problem: the insider is now the main attack surface. The Justice Department says North Korean operatives used American-run “laptop farms” to pose as remote employees at dozens of US companies; the FBI arrested an OpenAI employee accused of planting a hidden recording device in an executive conference room; and a startup called Pluto Security raised $42 million because employees are vibe-coding apps and connecting AI tools faster than security teams can see them. Technology leaders should inventory AI tools and agents, tighten identity checks for remote hires, and treat physical and device security as part of the same programme.
Security teams spent the past decade hardening the perimeter. This week’s news shows why that is no longer enough. In a single day, prosecutors described a state-backed scheme that put sanctioned workers on the payroll of US companies, an AI lab disclosed that an employee had allegedly hidden a recording device in its own office, and a funding announcement underlined how quickly employees are creating new, unmanaged software with AI. The common thread is trust: companies extend access to people and tools, and attackers, and sometimes honest staff, use that access in ways nobody planned for.
Story one: North Korean IT workers and the laptop farms
TechCrunch reported on October 13, 2026 that the US Department of Justice has warned of North Korean regime hackers increasingly relying on Americans to help break into US companies. Prosecutors said they had indicted and arrested at least two people and searched dozens of suspected “laptop farms” in Texas. Authorities seized more than 180 computers that US companies had shipped to new hires, not realising the workers were sanctioned North Koreans using stolen identities of US citizens. The DOJ said it raided over 40 laptop farms in Texas, the largest number of homes used for such schemes disclosed so far, owned by at least two American facilitators.
At least 70 US companies were infiltrated, according to prosecutors, with about $3 million in stolen funds. The companies were not named but were described as spanning healthcare, cyber support, retail, communications and manufacturing across Ohio, 16 other states and the District of Columbia. CrowdStrike, which helped the FBI, said laptop farms are “vital” to North Korea’s operators because they overcome the physical, financial and legal challenges of working from abroad.
The Justice Department said workers are recruiting American facilitators to counter improved private-sector detection, and in some cases subcontracting whole jobs to third-party foreign nationals. In a separate public service announcement, the FBI said there are “thousands” of North Korean state-controlled IT workers at Western companies and that they pose a significant insider-threat risk to national security and intellectual property. Cover identities include realistic email addresses, spoofed LinkedIn profiles and extensive social media histories.
Story two: a recording device in an AI lab’s boardroom
The second story is physical. OpenAI told authorities that an employee had allegedly planted a hidden recording device in an executive conference room at its San Francisco office. TechCrunch named the accused as Abdulrahman Alfozan, a Saudi national who joined OpenAI in July 2025 and worked on the product team. He was fired and arrested on October 9, about a week after the company discovered the device. Prosecutors with the US Attorney’s Office for the Northern District of California charged him with using an electronic device to intercept oral communications, which carries up to five years in prison. These are allegations, and neither he nor his attorney could be reached for comment in the report.
According to prosecutors, the device was inserted into a USB port on the back of a television. OpenAI had placed a camera there that allegedly caught the employee looking behind the TV. The company said it used its own AI models to understand the “sophisticated” device, which it described as having advanced encryption, audio monitoring and a data-transfer mechanism designed to evade monitoring, although it is unclear how technically distinct it is from other recording devices. OpenAI said it has no indication that the individual accessed or exfiltrated model weights, source code or customer data. FBI Director Kash Patel said the scheme aimed to “capture sensitive communications about the company’s super intelligence technology.”
The same day, TechCrunch reported that a federal judge ordered OpenAI to turn over hiring records in a trade secrets case brought by Apple, including interview materials involving former Apple employees. That case is separate, but it shows how closely hiring practices and trade secret protection are now linked.
Story three: employees are building the next attack surface
The third story is about software. Pluto Security emerged from stealth with $42 million in funding: an $8.5 million seed round led by TLV Partners and a $34 million Series A led by Jump Capital, with other investors including Skinos and Ansa Capital. CEO Shahar Bahat told TechCrunch that a newly announced AI tool can appear inside a company’s environment within two hours of release, and that about 60% of the risks the company detects across customers are in tools “vibe-coded” by non-developers. Finance departments are the biggest source, followed by marketing and legal; developers rank fourth. In one example, a finance employee at a Fortune 500 company built an app to manage customer information and churn and left it exposed to the internet.
Pluto began selling in April and says it has passed $2 million in annual recurring revenue with more than 30 enterprise customers, including Oscar Health, Workato, Sixt and AlphaSense. Bahat said, “there was not one POC in which we entered and didn’t find malicious stuff,” including extension packages and skills, some with active live attacks. Treat that claim as vendor-reported, but the direction matches what CIO.com has been publishing.
- Insider risk is multi-form: fake identities (North Korean workers), malicious employees (the alleged device case) and well-meaning staff creating unmanaged apps.
- Scale of the DOJ case: 40+ laptop farms raided in Texas, 180+ computers seized, at least 70 companies affected, about $3 million in stolen funds.
- AI sprawl is the new shadow IT: Pluto says about 60% of the risks it finds are in tools built by non-developers, with finance the biggest source.
- Governance is lagging: a CIO.com column argues AI governance models are “already obsolete,” and a Gartner report cited by CIO.com found 40% of C-level executives say AI is operational in some processes but only 13% report measurable, scaled results.
Trend context
Google’s US trending list on October 14 was dominated by entertainment and sport, but the strongest business signal was ASML, with more than 1,000 searches for “asml stock” after the chipmaker’s Q3 report (€11.7 billion in net sales, €5.0 billion in net income, order interest through 2030). The AI buildout is still being funded and equipped aggressively, which means more models, more agents and more tools landing inside companies. TechCrunch’s Disrupt coverage this week carried the same tension: executives such as Cerebras CEO Andrew Feldman argue the industry’s focus on risk drowns out benefits, while the security stories above show the risks are practical, not theoretical. For another recent view of how AI is shaping the technology agenda, see our note on Meta’s Muse Wonder launch.
What the CIO commentary adds
Recent CIO.com pieces give useful language for the problem. One argues that “your AI governance model is already obsolete,” because employees across every function now use provisioned AI tools to produce production-grade code, workflows and agents. Another urges organisations to measure AI abandonment as well as adoption, since licence counts show distribution, not value. A third, “AI is your newest hire. Manage it like one,” cites an August 2026 Gartner report in which 40% of C-level executives said their organisations had operationalised AI in some processes, but only 13% reported measurable results and scaling across the organisation. A separate column asks, of agents operating in shared channels, “whose authority is the agent spending?” Taken together, the message is that identity, authority and accountability need to apply to software agents as much as to people.
An action plan for technology leaders
1. Harden remote hiring and onboarding
- Verify identity with government documents and a live video check before day one, and again before granting privileged access.
- Ship laptops only to a verified home address; flag last-minute address changes, reshipments or requests to use a third-party address.
- Use endpoint management that detects remote-access tools, unusual input devices and geolocation or time-zone mismatches.
- Be cautious with contractors who subcontract work. Require contractual disclosure and approval of anyone else who touches your systems.
- Coordinate HR, security and IT. Our HR coverage discusses employment-law boundaries that should shape any new screening step.
2. Treat physical security as part of cybersecurity
The alleged device in the executive conference room was a USB-connected item behind a television. Basic controls would help many offices: disable or cover unused USB ports on shared equipment; inventory conference-room hardware; schedule periodic sweeps of sensitive rooms; restrict who can enter executive areas; and make it easy for staff to report unfamiliar devices. OpenAI’s use of a camera to observe the location is an example of detective control, which matters because the discovery came from monitoring, not from a policy alone.
3. Build an inventory of AI tools, agents and connectors
If tools can appear in your environment within two hours of release, quarterly reviews are too slow. Start with a discovery process: OAuth grants, browser extensions, AI coding assistants, plugins, MCP-style connectors and internal apps built with low-code or AI tools. Classify each by data sensitivity and owner. Give business teams a sanctioned way to build, such as a reviewed internal platform with logging and authentication, so that the secure route is also the easy route.
4. Give agents identities and limits
Treat an AI agent as a non-human identity with an owner, a purpose, minimum permissions, expiry dates and audit logs. Avoid shared credentials. Require explicit approval for agents that can send external messages, move money or change production systems. The CIO.com question about whose authority an agent is spending is a good test: if you cannot answer it for an agent, it should not be running.
5. Prepare for the legal and communications side
The OpenAI case includes a trade-secrets dispute in which a judge ordered the production of hiring records. Hiring files, interview notes and internal chats can become evidence. Work with counsel on retention policies, privilege and how you document investigations. Prepare an incident playbook that covers insider cases, not only external breaches: who can authorise an employee’s immediate access removal, how evidence is preserved, and when law enforcement is notified.
What not to overreact to
None of these stories justifies treating every employee as a suspect or banning AI tools. The alleged OpenAI device was found by the company’s own monitoring and, according to OpenAI, there is no indication that model weights, code or customer data were taken. The DOJ case involves specific, organised schemes. Pluto’s numbers come from a vendor with a commercial interest in finding problems. The reasonable response is proportionate: improve verification where the risk is highest, create safe pathways for building with AI, and monitor for abuse.
Bottom line
The perimeter was never the whole story, and this week made that plain. State-backed fraud, an alleged physical bug in a boardroom and unmanaged apps built by well-meaning staff are different threats with the same remedy: know who and what has access, verify before granting it, and watch for misuse afterwards. Technology leaders who can answer those questions for both people and AI agents will be better prepared for the next headline.
Sources
- TechCrunch, “Justice Department says North Koreans have infiltrated dozens of U.S. companies” (October 13, 2026)
- TechCrunch, “FBI arrests OpenAI employee accused of planting hidden recording device in conference room” (October 13, 2026); BBC News coverage of the same charge
- TechCrunch, “Judge orders OpenAI to turn over hiring records in Apple trade secrets case” (October 13, 2026)
- TechCrunch, “Pluto Security raises $42M as AI gives enterprise security teams a new headache” (October 13, 2026)
- CIO.com, “Your AI governance model is already obsolete”; “AI is your newest hire. Manage it like one”; “You’re measuring AI adoption. You should be measuring AI abandonment” (October 2026)
- Google Trends, US trending searches, October 14, 2026
Allegations described here are unproven. This article is general information, not legal or security advice.