The court calls Google’s bluff.

Google faces liability for AI-generated claims. Washington pauses public AI model assessments. Anthropic ships a safer AI model. OpenAI disrupts influence operations. Ransomware operators get a powerful new backdoor. Urgent patches land for Ivanti and Veeam. PyPI supply chain attacks evolve. And a massive data breach triggers a record fine in South Korea. Our guest is Peter Barker, Chief Product Officer at Ping Identity, sharing how identity increasingly becomes the control plane for how work gets done. AI analyzes the FIFA World cup, one cliché at a time.

Today is Thursday June 11th 2026. I’m Dave Bittner. And this is your CyberWire Intel Briefing.

A German court holds Google liable for AI output. 

A German court has ruled that Google can be held directly liable for false statements generated by its AI search overviews, marking a significant departure from legal protections traditionally granted to search engines. The Regional Court of Munich issued an injunction after Google’s AI falsely linked two Munich publishers to scams, subscription traps, and other dubious business practices. The court found that the AI had combined information from unrelated companies and created accusations that did not appear in any of the cited sources.

Central to the ruling is the court’s view that AI overviews are not merely search results. Unlike traditional search engines, which point users to third-party content, AI overviews generate new summaries, judgments, and conclusions. Because Google designs and controls the system, the court said it is responsible for those outputs as its own statements.

The court rejected Google’s argument that users can verify AI summaries by checking linked sources, noting that the summaries are presented as self-contained information and often contain claims not found in the sources. It also ruled that Google cannot rely on standard search-engine liability protections or Digital Services Act host-provider defenses.

The decision could have broad implications for AI providers. As generative systems increasingly create original summaries from web content, courts may hold operators accountable for inaccuracies, defamation, or unsupported claims produced by their models. Google was ordered to pay 80% of the legal costs, and the ruling may influence future cases involving AI-generated content worldwide.

The White House pumps the brakes on publishing AI model assessments. 

Trump administration officials have directed the Center for AI Standards and Innovation (CAISI), the federal government’s primary AI testing unit, to stop publishing public assessments of AI models while a new executive order is implemented. The move reflects growing concern over advanced AI systems, including Anthropic’s Mythos model, which officials worry could enable cyberattacks or support the development of biological weapons.

The order strengthens the role of national-security officials in AI oversight, a shift championed by National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent. While CAISI continues internal testing and coordination with government agencies, suspending its public reporting has raised concerns about its future and reduced transparency around AI risks.

The decision has exposed tensions within the administration and the AI industry. Companies such as OpenAI support preserving CAISI’s role, while others warn that stricter testing and security reviews could slow innovation and delay the deployment of advanced AI systems.

Anthropic releases Claude Fable 5. 

Anthropic has released Claude Fable 5, a new AI model derived from its more powerful Claude Mythos system, which the company previously restricted because of concerns it could help hackers identify and exploit software vulnerabilities. Fable includes additional safeguards designed to block responses related to cybersecurity, biology, and other sensitive topics, making it safer for broad public release.

Anthropic says most potentially risky requests will instead be handled by its earlier Claude Opus 4.8 model. The company argues these controls allow wider access while reducing security risks, though some researchers question whether such guardrails are fully reliable.

Mythos remains available only to a limited number of organizations responsible for critical infrastructure, helping them identify and patch vulnerabilities before attackers can exploit them. The debate highlights a growing divide over whether advanced AI capabilities should be tightly controlled or broadly shared to strengthen defensive cybersecurity research.

Still, some researchers remain unconvinced that Mythos represents a fundamentally new level of cyber capability, arguing that limited public access makes it difficult to determine whether the model’s reputation reflects a genuine breakthrough or effective marketing.

OpenAI says it has disrupted two China-linked influence campaigns. 

OpenAI says it has disrupted two China-linked influence campaigns that used ChatGPT to generate social media content aimed at shaping debate around U.S. technology and AI policy. One campaign promoted claims that AI data centers were driving up electricity costs, while another criticized U.S. tariffs and spread false allegations that ChatGPT user data had been compromised. OpenAI found no evidence that either operation significantly influenced public opinion. However, the company says the activity highlights how foreign influence actors are experimenting with AI-generated content to amplify existing political and economic concerns, target AI infrastructure debates, and attempt to manipulate public discussions while concealing their origins and motives.

MLTBackdoor supports ransomware operations by establishing a foothold on compromised networks. 

Researchers at Zscaler ThreatLabz have identified a new malware family called MLTBackdoor that appears designed to support ransomware operations by establishing a foothold on compromised networks. Delivered through a multi-stage ClickFix infection chain, the malware provides basic file-management capabilities but is particularly notable for its ability to load Beacon Object Files (BOFs), allowing operators to dynamically expand its functionality.

MLTBackdoor employs extensive obfuscation techniques, including Mixed Boolean-Arithmetic and Control Flow Flattening, along with anti-analysis measures that complicate reverse engineering and sandbox detection. It also uses indirect system calls, API hashing, encrypted communications, and a domain generation algorithm to maintain contact with command-and-control infrastructure.

According to ThreatLabz, the malware’s combination of stealth, resiliency, and modular BOF support makes it a capable post-exploitation framework that could facilitate lateral movement and other ransomware-related activity within victim environments.

Two critical vulnerabilities affect Ivanti Sentry. 

Ivanti has disclosed two critical vulnerabilities affecting Ivanti Sentry, including CVE-2026-10520, a remote unauthenticated command injection flaw that allows attackers to execute code as root, and CVE-2026-10523, an authentication bypass that enables the creation of rogue administrator accounts. A public proof-of-concept exploit for CVE-2026-10520 was released shortly after disclosure, increasing the likelihood of real-world attacks. Although Ivanti says it has not observed active exploitation, security researchers warn organizations should patch immediately. Fixed versions are available, and Rapid7 recommends updating affected systems outside normal patching cycles due to the severity of the flaws and the ease of exploitation.

Malicious PyPI package artifacts link to software supply chain campaigns. 

Researchers at Socket have identified 23 additional malicious PyPI package artifacts linked to the broader Mini Shai-Hulud, Miasma, and Hades software supply chain campaign, bringing the total known impact to 471 compromised packages across npm and PyPI. The latest wave shows attackers rapidly evolving their tactics, using a mix of Python startup hooks, trojanized native extensions, and new loader techniques to execute an obfuscated JavaScript stealer.

The malware targets developer workstations and CI/CD environments, seeking credentials, cloud secrets, package registry tokens, SSH keys, and other sensitive data. Researchers also observed anti-analysis techniques, including fake prompt-injection content designed to confuse AI-assisted security tools. According to Socket, the campaign demonstrates an increasingly sophisticated and adaptable threat that continues to shift delivery methods to evade detection and compromise software development ecosystems.

Veeam patches a critical remote code execution vulnerability. 

Veeam has patched a critical remote code execution vulnerability, CVE-2026-44963, affecting Backup & Replication version 12.x. The flaw, rated CVSS 9.4, allows an authenticated domain user to execute arbitrary commands on domain-joined backup servers. Successful exploitation could give attackers control over backup infrastructure, enabling them to delete, encrypt, or steal backup data, a common objective in ransomware attacks. The issue affects version 12.3.2.4465 and earlier 12.x builds but does not impact version 13. Organizations are urged to upgrade to version 12.3.2.4854 or later, implement hardened backup configurations, and restrict domain-user access while enforcing multi-factor authentication to reduce risk.

A data privacy breach costs South Korea’s largest online retailer $409 million. 

South Korea has imposed a record 624.7 billion won ($409 million) fine on Coupang, the country’s largest online retailer, over a massive data privacy breach and the unlawful collection of user information. Regulators said the company exposed personal data from 33 million customer accounts and 4 million nonmembers, while also improperly gathering online activity data from 11 million users across third-party websites and apps.

The Personal Information Protection Commission attributed the incident to inadequate security controls rather than sophisticated cyberattacks. Coupang, often called the “Amazon of South Korea,” has apologized and pledged to improve its data protection practices but plans to challenge the ruling in court.

The case has also become a diplomatic flashpoint, with some U.S. lawmakers accusing South Korea of unfairly targeting an American-incorporated company, while South Korean officials maintain the investigation followed standard legal procedures.

AI analyzes the FIFA World cup, one cliché at a time. 

With the FIFA World Cup kicking off in Mexico City, communications platform provider Sinch has launched the “Expected Cliché Tracker,” an AI-powered project that applies football’s love of analytics to an entirely different metric: manager press-conference platitudes. Inspired by expected goals, the tracker introduces “xC,” or expected clichés, a score designed to measure how often national team coaches rely on familiar football phrases instead of offering genuine insight.  

Once the tournament begins, the site will analyze every pre- and post-match press conference from all 48 national team managers, ranking the most original and least original speakers, tracking the tournament’s most-used clichés, and comparing coaches across countries and styles.  

The project promises daily storylines, head-to-head coaching comparisons, and a heat map of football’s favorite conversational habits. In other words, while teams compete for trophies on the pitch, managers will quietly compete for the far more elusive honor of avoiding phrases like “one game at a time.” 

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.