The breach wasn't in the model.

A cybersecurity startup claims to have accessed sensitive internal documents at Anthropic. The Pentagon spins up covert technology programs. AI developers get a crash course in key management. A ransomware recovery executive faces federal fraud charges. Two men are accused of using AI generated phone calls to scam bank customers. Stadia RAT rises from the StealC ashes. Researchers extract sensitive information using USB power-management signals. Patching what matters…matters. Monday business briefing. Tim Starks from CyberScoop is discussing targeted ads as a security risk. Championship rings come with a catch. 

Today is Monday October 26th 2026. I’m Dave Bittner. And this is your CyberWire Intel Briefing.

A cybersecurity startup claims to have accessed sensitive internal documents at Anthropic. 

Cybersecurity startup Cathedral says its researchers used AI agents to gain access to sensitive internal systems at Anthropic in less than twelve hours. According to The Washington Post, the researchers exploited vulnerabilities in third-party business software, accessing systems containing meeting transcripts, salary information, and internal Slack messages.

Anthropic confirmed the intrusion but said access was limited to certain sales, human resources, and IT systems. Customer conversations, proprietary code, and AI model weights were not exposed. The company says its monitoring detected some suspicious activity, and it disabled the affected systems and worked with vendors to address the vulnerabilities.

Cathedral, founded by former DOGE employees, conducted the research through Anthropic’s security testing program. Anthropic disputes the startup’s characterization of the national security implications and says researchers violated some program rules.

The findings highlight a familiar security problem with potentially higher stakes: even cutting-edge AI companies depend on ordinary enterprise software, with all its ordinary vulnerabilities.

The Pentagon spins up covert technology programs. 

The Pentagon has quietly established a new agency dedicated to disrupting foreign military technology programs through covert operations.

According to POLITICO, the Technology Support Office, established in April, consolidates two classified intelligence programs and aims to expand capabilities including cyberattacks, supply-chain sabotage, and deception operations.

The agency reportedly employs engineers, intelligence officers, and special operations personnel to develop tools that could delay adversaries’ technological advances, particularly in artificial intelligence and advanced weapons systems.

Deputy Defense Secretary Steve Feinberg spearheaded the initiative, which has attracted opposition from Cyber Command and Special Operations Command over competing responsibilities and resources.

Supporters argue the effort is necessary to preserve America’s military advantage. Critics worry its secrecy could enable questionable operations and undermine congressional oversight.

The Pentagon disputes POLITICO’s reporting, calling its underlying claims inaccurate.

AI developers get a crash course in key management. 

A supply-chain attack affecting tens of thousands of AI developers began with a security mistake made months earlier.

LiteLLM says it accidentally published a configuration file containing a master key in May. Although the file was removed, the key was never revoked. An attacker subsequently used it to gain control of LiteLLM’s GitHub repositories and distribute credential-stealing malware through a compromised software update.

Security firm Wiz says the malicious package was downloaded roughly 49,000 times. Within hours, attackers were testing stolen AWS credentials for access to AI models, suggesting a possible LLMjacking operation.

LiteLLM warns anyone who installed version 1.107.0 to assume their credentials were stolen and rotate them.

No organization has publicly confirmed a resulting breach. It’s LiteLLM’s second supply-chain compromise this year.

A ransomware recovery executive faces federal fraud charges. 

A Florida businessman faces federal fraud charges for allegedly charging ransomware victims millions of dollars to recover their data while secretly paying the hackers responsible.

Prosecutors say Zohar Pinhasi, who operated the data recovery firm MonsterCloud, claimed he could decrypt files without paying ransoms. Instead, investigators allege he collected more than $19 million from clients while paying hackers over $8 million. In one case, an $8,200 ransom payment reportedly became a $150,000 customer bill.

The indictment follows a 2019 ProPublica investigation exposing the alleged scheme. MonsterCloud’s clients included law enforcement agencies that believed they were avoiding payments to criminals.

Pinhasi has pleaded not guilty and faces up to 20 years in prison if convicted.

For victims already paying one extortionist, prosecutors allege MonsterCloud effectively added another.

Two men are accused of using AI generated phone calls to scam bank customers. 

Dutch police have arrested two men accused of using AI-generated phone calls to steal more than $425,000 from 170 bank customers.

The scheme began with automated calls warning of suspicious account activity. Victims who pressed a button were connected to a fake bank employee, who instructed them to download an app and scan their debit cards.

The app captured contactless payment data, which accomplices used to withdraw cash from ATMs.

The suspects, ages 23 and 25, were arrested October 6. Police also seized servers, phones, and laptops.

Both remain in custody as the investigation continues.

Stadia RAT rises from the StealC ashes. 

Four months after European law enforcement disrupted the StealC malware operation, its seller is back with a new product.

Researchers at Varonis say Stadia RAT builds on StealC’s credential-stealing capabilities, adding remote desktop control, command execution, file management, and the ability to deploy additional malware.

The Windows malware targets browsers, cryptocurrency wallets, and messaging applications. It also avoids systems configured for several Eastern European and Central Asian languages, establishes persistence through scheduled tasks, and can disable portions of Microsoft Defender’s protection.

Subscriptions range from $300 to $900 a month, with former StealC customers offered a free trial.

Researchers confirmed that Stadia reuses code from StealC, illustrating a familiar limitation of law enforcement takedowns: dismantling criminal infrastructure doesn’t necessarily put its operators out of business.

Researchers extract sensitive information using USB power-management signals. 

Researchers at Korea University have demonstrated a technique for extracting sensitive information from air-gapped computers using USB power-management signals.

Called SPECTER, the method allows malware with administrator privileges to encode data through changes in USB device states, without conventional network connections or USB data transfers.

Researchers demonstrated three ways to receive the signals: monitoring smartphone charging indicators, observing changes in USB peripherals, and detecting electromagnetic emissions from USB cables.

Transmission speeds were extremely slow, topping out at one bit per second, making the technique more suitable for stealing passwords or cryptographic keys than large files.

The attack also requires hardware or firmware modifications, and some configurations produced no usable signals.

Researchers suggest monitoring unusual USB power-management activity, although the effectiveness of that approach remains untested.

Patching what matters…matters. 

Security teams are patching more vulnerabilities than ever, but new research suggests all that effort isn’t necessarily making organizations safer.

According to Qualys, vulnerability closures increased sixfold between 2022 and 2025, reaching 473 million. Yet the percentage of critical vulnerabilities remaining unpatched after seven days actually increased.

The problem? Prioritization.

Researchers say fewer than one percent of vulnerabilities are actively exploited, but organizations continue allocating resources based largely on severity scores rather than evidence of real-world attacks.

Meanwhile, attackers are moving faster. Across the most critical exploited vulnerabilities, manual remediation failed to keep pace 88 percent of the time.

Qualys recommends shifting toward intelligence-driven, automated remediation that prioritizes actual exploitation and business impact.

The takeaway: Security isn’t necessarily about patching more. It’s about patching what matters.

Monday business briefing. 

Investment and acquisition activity remains brisk across cybersecurity, with AI security attracting particular attention.

San Francisco-based Reach Security raised $50 million in Series B funding to expand its exposure management platform. AI agent security startup Opsin secured $21 million, while Asymptote Labs raised $5.5 million for its AI runtime security technology. AI penetration testing firm ManticoreAI also announced an undisclosed seed round.

On the acquisition front, Proofpoint agreed to buy data and AI security company Varonis for $6.3 billion, potentially creating a combined business serving more than 16,000 enterprises.

Anthropic acquired confidential AI data specialist OPAQUE, while Canadian AI company Cohere agreed to purchase Czech vulnerability management provider AISLE.

Elsewhere, legal technology firm Lineal acquired Swiss digital forensics company Forentec, and Britain’s Bytes Technology Group purchased Microsoft security specialist CyberOne.

The common thread across these deals is growing investment in AI security, data protection, and automated vulnerability management.

 

Championship rings come with a catch. 

The New York Knicks celebrated their first NBA championship in more than half a century by offering employees free championship rings. Unfortunately, the offer was fake, and the real prize was mandatory cybersecurity training.

According to The New York Times, Madison Square Garden sent employees a simulated phishing email inviting them to choose between two ring designs. Anyone who clicked the link earned a trip to cybersecurity class instead.

The exercise followed a real security incident in which attackers reportedly used phished employee credentials to hack and extort the organization.

MSG acknowledged that using championship rings as bait was an unfortunate mistake. The company says it’s distributing more than 275 actual rings and pieces of jewelry worth over $3.7 million.

For employees who fell for the phishing test, the championship celebration apparently came with a refresher course in disappointment.

And that’s the CyberWire.

For links to all of today’s stories, check out our Daily Briefing at the cyberwire dot com.

 

We’d love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like the show, please share a rating and review in your podcast app. Please also fill out the survey in the show notes or send an email to cyberwire@n2k.com

We’re proud that N2K CyberWire is part of the daily routine of the most influential leaders and operators in the public and private sector, from the Fortune 500 to many of the world’s preeminent intelligence and law enforcement agencies.

N2K helps cybersecurity professionals and organizations grow, learn, and stay ahead. We’re the nexus for discovering the people, tech, and ideas shaping the industry.  Learn how at n2k.com.

 

N2K’s lead producer is Liz Stokes. We’re mixed by  Tré Hester, with original music by and sound design Elliott Peltzman. Our contributing host is Maria Varmazis. Our executive producer is Jennifer Eiben. Peter Kilpe is our publisher. And I’m Dave Bittner. Thanks for listening.