A little too OpenAI.

Dave Fittner reports that 35-year-old OpenAI engineer Abdurrahman Al-Fozan, a Saudi national employed by OpenAI since July 2025, was arrested and charged with allegedly placing an encrypted recording device in an executive conference room at the company’s San Francisco offices. Prosecutors say the device was hidden inside a television USB port and wrapped in toilet paper. Investigators reportedly found chatbot queries tied to Al-Fozan’s spouse’s account containing instructions for building and concealing audio recorders, and security staff later found a second device in his unattended bag. The room hosted discussions about proprietary research, product development, and government relationships; authorities have not said whether anything was successfully recorded. Al-Fozan faces up to five years in prison if convicted. The briefing then warns that attackers are exploiting a high-severity Palo Alto Networks PAN-OS vulnerability affecting PA-Series and VM-Series firewalls and Panorama management systems. Authenticated attackers with network access can execute code with root privileges. CISA ordered federal agencies to patch and investigate by October 17, while Palo Alto said exploitation was limited. Because patching is not enough, defenders are told to review administrative logins, investigate suspicious configuration changes, and restrict management interfaces to trusted networks. CISA also urged immediate action on an Ivanti Connect Secure zero-day that permits authentication bypass and remote command execution. Veloxity observed exploitation beginning October 3, including credential theft, searches for administrator passwords and BitLocker keys, Active Directory probing, log deletion, and memory manipulation intended to evade Ivanti’s integrity checker. Ivanti released patches October 13, but another privilege-escalation issue remained unpatched; organizations were told to rebuild compromised appliances and rotate exposed credentials and certificates.

Symantec says the relatively new Aoki infostealer has expanded beyond browsers, cryptocurrency wallets, and password managers. Newer versions seek authentication tokens, session files, and API keys connected to Claude, ChatGPT, Codex, GitHub Copilot, Cursor, and Gemini. The operators have not been identified, but Aoki’s continuing focus on cryptocurrency suggests a financially motivated group, and the malware remains under active development. The briefing explains why AI credentials are valuable: attackers could impersonate developers, generate unauthorized API charges, or use the stolen access to reach code repositories, cloud services, and development pipelines. The report does not establish that Aoki has accessed all of those systems, but presents them as potential consequences of the credential theft. Checkpoint researchers also found two high-risk vulnerabilities in LightLLM, an open-source gateway connecting enterprise AI applications with providers including OpenAI, Google, and Anthropic. An ordinary authenticated user could potentially exploit a vulnerable deployment. One flaw enables remote code execution and possibly complete system takeover; the other can expose API keys through manipulated outbound requests. Because AI gateways often centralize credentials, permissions, and access to several providers, a compromise could have an unusually broad effect. The affected versions have fixes available, and organizations are urged to update promptly.

Darktrace researchers uncovered a campaign distributing ChikabikaBot, a Rust-based remote-access trojan that uses Discord for command and control. The infection begins with a document-themed lure that launches hidden PowerShell commands and attempts to bypass Windows security protections. Once installed, the malware gathers system information, captures screenshots, executes commands, manipulates files, and supports surveillance. It uses ngrok to exfiltrate stolen data, illustrating the campaign’s broader lesson: legitimate online services can provide infrastructure for malicious operations. Citizen Lab researchers at the University of Toronto described a separate surveillance operation called AdFection, in which ordinary online advertisements deliver mercenary spyware without requiring the victim to click. They allege that Israeli advertising company DiscoverTech facilitated surveillance for spyware vendors including Candiru. Citizen Lab linked the activity to zero-click attacks against at least 45 devices at Al Jazeera. DiscoverTech advertisements appeared on more than 120,000 websites and 70,000 mobile applications between March and June, with related activity seen on sensitive government and international networks. Researchers also identified “Dark Cartographer,” an allegedly related operation that exploits mobile-carrier weaknesses to connect advertising identifiers with phone numbers. Google and IAB Europe removed DiscoverTech from approved vendor lists. Citizen Lab’s conclusion is that the consumer-targeting machinery of digital advertising can be repurposed to target people for surveillance, even without an intentional interaction.

Interpol says a coordinated operation across 17 countries in Latin America and Europe produced 95 arrests related to the distribution of child sexual abuse material through online messaging groups. International Effort Confluencia combined four investigations conducted between March and September, and authorities executed 127 search warrants. The operation included several specific cases. Investigators rescued a five-year-old child whose parent was allegedly producing and selling abuse material. Police in Panama arrested the administrator of a messaging group with more than 700 members. In Argentina, a suspect allegedly used sextortion against girls in multiple countries. Interpol warned that messaging platforms are increasingly being used for grooming, exploitation, and commercial distribution of abuse material. Authorities expect the evidence seized in the searches to help identify further offenders and children at risk. The briefing presents the operation as ongoing investigative work rather than a completed accounting of the networks involved.

The featured Hacking Humans segment has Joe Carrigan discuss a Wisconsin Watch report on cryptocurrency ATMs and the people profiting from the scam marketplace. Tennessee, Minnesota, and Indiana banned crypto ATMs outright, while Wisconsin adopted Act 226. The law limits daily transactions to $1,000, requires receipts, mandates customer identification, and provides a refund process for victims. Wisconsin Department of Financial Institutions legal counsel Michael Gaffigan said reports of illicit activity had fallen, and Greenville police cryptocurrency investigator Scott Simmons said ID requirements make it harder for fraudsters to use multiple people’s identities. The hosts cite a real example: 71-year-old Karen Schmeling lost $4,400 to a crypto-ATM scam in 2023 and said the operator refused a refund despite her submitting extensive, notarized paperwork. Under the new law, a victim who contacts the ATM operator and the Department of Justice, Department of Financial Institutions, or law enforcement within 30 days can seek recovery. Joe notes that reducing a single-day loss from $20,000 to $1,000 is still a major improvement. The segment says the financial evidence is extreme. An Iowa attorney general’s investigation found that more than 98% of money sent through Bitcoin Depot and almost 95% sent through CoinFlip were scam transactions. Bitcoin Depot later filed for bankruptcy, citing restrictive legislation. Digital Mint co-founder Mark Grenz said nearly all customers making transactions of about $1,000 or more were fraud victims; his company voluntarily shut down after operating about 1,400 locations. Fees can reach 30%, making online exchanges more practical for legitimate users. Kieran Heumann adds that Bitcoin itself is traceable transaction by transaction, but funds can be exchanged into other cryptocurrencies and moved abroad within roughly 30 minutes, making recovery difficult.

The Hacking Humans discussion turns to a legal dispute involving the Wood County Sheriff’s Office in Wisconsin. Beginning in March 2023, the county obtained court permission to search and eventually seize money from crypto ATMs when a user reported being defrauded. The approach spread to other Wisconsin agencies, but CoinHub sued Wood County in June 2024, arguing that officials had skipped required legal steps before taking cash from its kiosks. Joe says he supports stopping scams but does not support police simply taking an ATM and opening it because they believe the cash represents ill-gotten gains. He gives a hypothetical warning: someone could put $5,000 into an ATM, receive cryptocurrency worth roughly $3,000 after fees, falsely claim the machine stole the money, and potentially obtain both the returned cash and the cryptocurrency. His conclusion is that the legal process must be followed even when officers believe they are helping victims. Heumann agrees that the issue is difficult because cryptocurrency law often lags behind the technology. He explains that Bitcoin transactions are highly traceable, but users can quickly exchange Bitcoin for less-traceable cryptocurrencies and move the funds internationally. That may explain why officers try to recover cash immediately, but Heumann says there is no simple right answer. The hosts ultimately treat Wisconsin’s consumer protections as useful while separating them from potentially improper seizure practices.

The episode closes by presenting Character AI’s bankruptcy as a warning for the broader AI industry. The company was valued at $1 billion in 2023 and attracted millions of users, but computing costs reached as much as $13.8 million per month while subscription revenue never exceeded $2.8 million. Lawsuits alleging harm to users added significant legal expenses, although Character AI acknowledged that its business was unsustainable even without those costs. The company tried to reduce the gap by simplifying its models, laying off employees, introducing advertising, and charging for additional features. Those measures diminished the user experience without solving the financial problem: Character AI still lost approximately $8 million to $9 million per month. The service remained operational during bankruptcy proceedings. The briefing contrasts Character AI with larger competitors that can rely on deep-pocketed investors or established businesses to subsidize AI development. Its lesson is that impressive technology and enthusiastic users do not automatically produce a viable company when each interaction carries substantial computing expense and revenue does not cover it.