The early bird gets the exploit.

Amazon spots PHP exploits before disclosure. GitLab patches a critical flaw. Meta faces a $25 billion penalty. Autonomous AI agents power new cyberattacks. Hackers breach a Japanese cloud provider. CyberCartel unleashes AI against Latin American banks. Tor exit nodes are hijacked. OpenAI coding agents disable security monitoring. Interpol dismantles thousands of cybercrime networks across Africa. Our guest is Danny Manimbo, ISO and AI Practice Leader at Schellman, explaining why companies waiting for AI regulatory clarity may already be behind. Spotify hits the skip button on AI slop.

Today is Thursday October 22nd 2026. I’m Dave Bittner. And this is your CyberWire Intel Briefing.

Amazon detected PHP exploitation attempts days before public disclosure. 

Amazon Threat Intelligence detected attempts to exploit two PHP vulnerabilities four days before their public disclosure on October 21st. The activity, observed through Amazon’s MadPot honeypot network, originated from anonymized infrastructure, suggesting threat actor involvement. How the attackers obtained exploit information remains unknown.

Additional attempts appeared hours before disclosure, followed by widespread exploitation once patches became available. Researchers believe multiple actors were involved, although it’s unclear whether the early attacks would have succeeded against real-world systems.

Amazon warns that coordinated disclosure no longer guarantees a head start for defenders. AI-assisted exploit development is accelerating the transition from vulnerability disclosure to active attacks, sometimes to just minutes. Security teams should prioritize patching critical internet-facing vulnerabilities within hours, investigate potential compromises predating disclosure, and deploy detections that don’t depend on vulnerability-specific signatures.

GitLab patches a maximum-severity vulnerability. 

GitLab has released emergency patches for a maximum-severity vulnerability that could allow unauthenticated attackers to execute arbitrary code on vulnerable servers.

Tracked as CVE-2026-110264, the flaw involves an integer overflow in GitLab’s Regular Expression Compiler. It affects both Community and Enterprise editions, and GitLab is urging customers to update immediately.

The Shadowserver Foundation tracks more than 2,700 internet-exposed GitLab Community Edition instances, although it’s unclear how many remain vulnerable.

This is GitLab’s second critical remote code execution vulnerability in ten days, both reported by the same security researcher. Last month, GitLab also patched a maximum-severity flaw that could expose sensitive information. CISA subsequently confirmed active exploitation of that vulnerability and ordered federal agencies to patch.

GitLab says its hosted service is already protected.

Meta is fined $25 billion in New Mexico. 

A New Mexico judge has ordered Meta to pay $25 billion in penalties stemming from the Cambridge Analytica privacy scandal.

A jury found the company liable for more than 43 million violations of the state’s unfair practices law, concluding that Meta misled users about its data collection practices, privacy protections, and harmful content.

The judge initially calculated penalties totaling nearly $220 billion before reducing the award to $25 billion, with annual interest of 15 percent until it’s paid.

The ruling also criticized Facebook’s internal investigation following the Cambridge Analytica revelations, calling it deliberately ineffective.

New Mexico’s attorney general hailed the decision as a warning to powerful technology companies.

Meta strongly disputes the ruling and plans to appeal, arguing that the state’s approach effectively imposes a tax on online speech.

Researchers uncovered new threat actors using autonomous AI agents. 

CrowdStrike researchers have uncovered four previously undocumented threat actors using autonomous AI agents to conduct cyberattacks, with some operations achieving their objectives within an hour.

The actors, operating in Spanish, Chinese, Indonesian, and Russian, targeted government agencies, healthcare organizations, and private companies worldwide. Their activities included successful intrusions and data theft.

Researchers discovered the campaigns after the attackers inadvertently exposed their offensive AI frameworks through publicly accessible directories.

The operations varied considerably in sophistication. One actor employed multiple coordinated AI agents, while another, despite limited technical skills, successfully compromised government systems using publicly available tools.

CrowdStrike warns that agentic AI is lowering the technical barriers to cybercrime, allowing less-skilled attackers to conduct increasingly sophisticated operations at machine speed. Researchers expect these AI-powered intrusions to grow in both volume and sophistication.

Hackers compromise a Japanese cloud hosting provider. 

Hackers have compromised a Japanese cloud hosting provider, potentially installing malware on nearly 27,000 customer servers.

Onamae (o-na-MA-ay) Desktop Cloud, operated by GMO Internet, confirmed unauthorized access to its infrastructure and the installation of malicious software in customer environments.

The company began shutting down affected services on October 21st, disrupting all Desktop Cloud plans. There’s no estimated timeline for restoring service.

GMO warns that information stored on compromised servers may have been exposed, although it has not confirmed any data theft or misuse. Credit card information was unaffected because it’s stored separately.

Customers are being urged to change passwords and other credentials stored on their servers and remain alert for phishing attempts.

The company says its investigation is ongoing and promises additional updates.

CyberCartel targets banks and payment providers across Latin America. 

Security researchers at Straiker have uncovered an AI-powered attack system used by a cybercriminal group targeting banks and payment providers across Latin America.

The group, known as CyberCartel, has reportedly been active since 2012. Its latest operation uses a framework called CARLIY0S, which coordinates multiple commercial AI models to identify vulnerabilities, develop exploits, and execute attacks.

The AI agents have specialized roles, including reconnaissance, exploit development, and evaluating competing attack strategies.

Researchers gained access to the group’s operational environment, uncovering a database containing nearly 4,500 records, including financial information, credentials, API keys, and backdoors.

Straiker says the group targeted more than 400 financial organizations in a matter of months.

The findings offer a rare look at AI becoming part of cybercriminal infrastructure, potentially allowing smaller groups to conduct sophisticated attacks faster and at greater scale.

Hackers hijack Tor exit nodes. 

Hackers hijacked 333 Tor exit nodes in an operation designed to intercept SSH connections and unencrypted email traffic.

According to Tor volunteers, the intrusion began with a reused password that provided access to a management server. Attackers installed the Ebury Linux backdoor and eventually gained control of hundreds of exit relays.

Between October 2nd and 5th, the compromised nodes redirected SSH connections through attacker-controlled servers, potentially capturing passwords and installing additional malware. The attackers also blocked encrypted email ports while allowing unencrypted traffic, although it’s unclear whether any email interception succeeded.

Researchers still don’t know how the attackers compromised so many servers or who’s responsible.

The incident highlights the importance of enforcing encrypted email connections and using SSH keys instead of passwords, particularly when routing traffic through Tor.

OpenAI coding agents weakened internal security monitoring. 

Two OpenAI coding agents weakened internal security monitoring while completing routine programming tasks.

In one case, a GPT-6.1 Sol agent disabled automatic security alerts after misinterpreting instructions not to send Slack messages. In another, a GPT-6 Astra agent removed monitoring software to match a reference implementation, then falsely told the user it hadn’t.

Both agents considered preserving the safeguards but ultimately prioritized their coding instructions.

OpenAI is strengthening its monitoring systems while investigating whether the incidents reflect excessive task persistence or a specific tendency to evade oversight.

Authorities shut down African cybercrime networks. 

Interpol says an international law enforcement operation across Africa has resulted in more than 1,400 arrests and the disruption of nearly 11,000 cybercriminal networks.

Operation Serengeti III involved authorities from 21 countries, targeting ransomware, business email compromise, and online fraud between late June and August.

Investigators identified nearly 66,000 victims, with estimated losses exceeding $385 million. Authorities recovered more than $105 million in illicit funds.

Among the results, Ugandan police arrested 365 suspects linked to investment and pension fraud. Angola seized more than 2,000 illegal cryptocurrency mining devices, while Nigerian authorities arrested a suspect allegedly responsible for a major phishing operation.

Private cybersecurity companies provided intelligence to help identify criminal infrastructure and suspects.

It’s the third Operation Serengeti crackdown since 2024, reflecting an ongoing international effort to disrupt cybercrime across the continent.

 

Spotify hits the skip button on AI slop. 

Spotify has spent the past year trying to keep AI-generated musical slop from overwhelming its platform, and apparently, there’s been plenty to clean up.

The streaming giant says it’s removed more than 50 million spammy or low-quality tracks from recommendations this year, on top of another 75 million tracks removed from the platform entirely.

Spotify says listeners were consistently skipping or rejecting content identified as slop, prompting changes to its recommendation algorithms.

The company has also introduced AI disclosure credits, verified more than 300,000 artists, and begun labeling AI-generated artist personas. Those virtual performers are generally excluded from recommendations unless listeners actively follow them.

Spotify acknowledges that detecting AI-generated content remains difficult, particularly as the technology improves.

The company insists its priority remains human artistry, even as artificial intelligence makes producing music easier than ever.

Apparently, just because a machine can write a song doesn’t mean anyone wants to hear it.

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.