Crypto's biggest withdrawal.

The briefing opens with Binance’s reported $6.3 billion cryptocurrency theft on October 29, described as possibly the largest crypto hack on record. TRM Labs and Chainalysis attributed the attack to North Korea’s Trader Trader operation. The attackers moved funds across 10 blockchains, with most of the theft occurring within roughly 20 minutes. Binance’s preliminary investigation points to a compromised DevOps engineer’s laptop, which may have opened access to systems authorized to move funds. The company found no evidence that private wallet keys were stolen. Instead, the attackers apparently manipulated transaction-signing infrastructure, including an Ethereum feature that automated withdrawals. Issuers and networks froze billions of dollars in assets, but estimates still put the attackers’ control at $2.4 billion to $3.3 billion. Binance promised to reimburse affected customers, while the incident leaves major questions about the controls protecting exchange wallets. The same briefing also reports a temporary Polymarket trading suspension after a cyberattack through third-party provider DebugBear. Polymarket said trading resumed and that affected users would be reimbursed, but it did not disclose total losses or name the attacker. Researchers traced more than $2.1 million in suspected crypto theft to one account and found malicious code apparently designed to identify wallets and transfer their assets. Polymarket removed DebugBear’s embedded code; DebugBear acknowledged an incident and urged customers to remove its code, while researchers found no evidence that other customers were compromised.

Attackers are actively exploiting a maximum-severity GitLab vulnerability that can enable unauthenticated remote code execution. The flaw is an integer overflow in GitLab’s regular-expression compiler, allowing malicious code to be executed through specially crafted CI/CD configurations. CISA added it to the Known Exploited Vulnerabilities catalog and directed federal agencies to secure affected systems by Sunday. GitLab patched the flaw earlier in the month and issued additional threat detections after observing exploitation attempts. The Shadowserver Foundation counted nearly 2,800 Internet-exposed GitLab Community Edition instances, although the briefing says it is unclear how many remain vulnerable. The incident follows other serious GitLab problems this year, including a recently exploited issue that could expose credentials and other secrets. The briefing then turns to a covert Russian influence operation targeting Texas Democratic Senate candidate James Tallarico with fabricated misconduct allegations. Aletheia and Clemson University analysts attributed it to Storm 1516, a group associated with similar false claims during the 2024 presidential election. Researchers at Graphika believed the videos used in the campaign were AI-generated. One video gained more than 850,000 views in its first 24 hours after influential accounts on ex-Twitter amplified it. U.S. intelligence says Russia has intensified election interference, especially in competitive Senate races, but officials have not concluded that Moscow is explicitly favoring either party.

The U.S. Treasury Department sanctioned Russian and North Korean people and companies accused of supporting Pyongyang’s fraudulent remote IT-worker operations. Two Russians allegedly laundered millions of dollars in cryptocurrency earned by North Korean operatives working for American companies. Treasury also sanctioned North Korean managers overseeing workers in Russia and a company accused of using artificial intelligence to help operatives obtain U.S. jobs. The broader scheme relies on stolen identities to place North Korean workers in lucrative technology positions. The workers send their earnings back to Pyongyang; some have also stolen sensitive information or extorted employers. U.S. officials say the resulting revenue supports North Korea’s weapons programs. The sanctions followed a Justice Department operation targeting laptop farms used to make overseas workers appear to be located in the United States. The episode also reports that Britain’s GCHQ cannot account for thousands of storage devices, including USB drives, hard disks, and memory cards. The problem was identified years earlier, but an investigatory powers commissioner criticized the agency’s slow investigation. Some missing devices could contain intercepted communications and highly sensitive intelligence data. GCHQ acknowledged shortcomings in its tracking system and said it had introduced corrective measures, but officials were still determining how many devices were actually lost.

Researchers at iResearch found that an abandoned version of the Popcorn Time streaming application, called Time for Popcorn, left hundreds of thousands of Macs potentially exposed. The app installed a persistent update checker capable of downloading and executing code with elevated privileges even years after users stopped opening it. Because the developers had disappeared, attackers could potentially take over expired update domains and use them to distribute malware. The researchers found no evidence of exploitation. They estimated that 250,000 to 350,000 Macs were affected, then took control of an abandoned domain and used the update mechanism to disable the vulnerable checker on more than a million Macs and Windows PCs. Other vulnerabilities remain, and some update domains are still uncontrolled, so the recommendation was to uninstall Time for Popcorn entirely. Japan Airlines disclosed that a breach at its technology provider, Amadeus IT Group, exposed information belonging to 92,000 customers. The affected data included names, reservation codes, and flight details. For about 2,000 passengers, attackers also obtained passport numbers, contact information, and frequent-flyer numbers. Japan Airlines said payment-card information and flight operations were unaffected. The disclosure came amid other Japanese incidents involving a Sony-owned internet provider, Burger King Japan, and Capcom, prompting Prime Minister Sanei Takaichi to call for emergency cybersecurity measures and increased concern about stolen personal information.

Cisco Talos researchers discovered an AI-integrated Windows command-and-control framework called NIC-C2 operating in the wild. The malware combines credential theft, keylogging, surveillance, remote access, lateral movement, persistence, and security evasion with direct access to OpenAI’s API. Attackers can use AI during an intrusion to analyze stolen information and potentially speed up operational decisions. NIC-C2 can also use a victim’s own OpenAI API key, potentially charging the victim for the attacker’s requests. Its ransomware component locks the screen and demands cryptocurrency without encrypting files. Talos observed active infrastructure and multiple versions of the malware, which the researchers interpreted as evidence of continued development. The briefing’s warning is that putting AI directly into malware could reduce defenders’ time to detect and interrupt an attack. The framework is not presented merely as a chatbot attached to malware; it can support analysis and control during an ongoing intrusion. That combination of surveillance, credential theft, remote access, and AI-assisted decision-making makes the malware an example of how attackers may use models to accelerate existing attack workflows.

Host Dave Bittner interviews Josh Devon, co-founder and CEO of Sondera, moments after the company won the ninth annual Data Tribe Challenge. Devon says the win was gratifying because Sondera operates in a crowded space where it can easily be lumped together with competitors. The company’s central problem is controlling AI agents with deterministic rules. Devon argues that agents cannot simply be trusted to follow instructions, and that each organization—not a vendor—must define the rules its agents should obey. Sondera’s approach lets users express those rules in natural language and then converts them into verifiable policy code. Devon calls this process “auto-formalization.” During the previous 15 or 16 months, the company had published research on the subject at venues including ICML and FLoC while building technology to enforce the resulting policies. He describes agents as black boxes, comparing their internal state to the unknowability of another person’s thoughts. The goal is therefore not only to prohibit unwanted behavior but also to specify positive obligations describing how an agent must act. Bittner asks whether monitoring agents will destroy their speed advantage. Devon says Sondera uses lightweight sensors, neural classifiers, and, where needed, LLMs as judges. Benchmarking showed that the system adds only a couple of seconds, and the underlying model driving the agent is usually the larger bottleneck—for example, a model may already take 13 seconds to make a tool call. The company’s main customers are regulated enterprises that want agents to handle sensitive information or move money but cannot accept the risk of accidental data leakage or unauthorized transfers. Devon says the Data Tribe Challenge provided exposure, a way to stress-test the technology, and a forced opportunity to distinguish Sondera from roughly 27 other companies pitching agent control. He describes Sondera as a frontier research lab that publishes papers and open-sources some technology so others can test its claims. His comparison is to encryption: organizations should not merely trust it; they should be able to prove it works. Sondera’s next priority is finding customers who want to deploy agents in higher-ROI, higher-risk workflows.

The episode closes by saying goodbye to Sir Patrick Stewart, reported as having died at 86. The host recounts Stewart’s path from a difficult working-class childhood in Yorkshire to the Royal Shakespeare Company and, approaching age 50, to the science-fiction television series that made him Captain Jean-Luc Picard. Picard is remembered as thoughtful, principled, intellectually curious, diplomatic, and committed to human dignity. The host emphasizes that Picard listened, questioned his assumptions, and understood that having authority to give an order did not automatically make it the right order. The tribute connects Star Trek to generations of scientists, engineers, programmers, and technology professionals. Communicators, tablets, voice-controlled computers, and networks connecting people across great distances are cited as technologies that the show helped audiences imagine and, in some cases, later build. Stewart helped make that future feel worth pursuing and modeled the idea that intelligence and kindness, leadership and compassion, can coexist. The cybersecurity connection is Picard’s early use of multi-factor authentication: activating or canceling the Enterprise’s auto-destruct sequence required authorization codes and confirmation from multiple senior officers. As the host puts it, even in the 24th century, “you couldn’t just blow up the Enterprise with a single password.” The episode ends with Picard’s reflections on time, mortality, and how people live, followed by notices for Research Saturday and Sunday’s T-Minus Space Cyber Briefing.