Authorities dismantle KillSec group

0:03Authorities dismantle the Killsec hacking group. South Africa thwarts a ransomware attack against air traffic control systems. The US sanctions Venezuelan ATM hackers. And a Chinese APT targets AI experts. This is the Risky Bulletin, prepared by Katalin Kimpanu and read by me, Claire Aird. Today is the 2nd of October, and this podcast episode is brought to you by Port Swigga. In today's top story, European law enforcement agencies have arrested three members of the Kilsec ransomware group and seized their servers. The group's main administrator, a 16-year-old Romanian, was arrested in Spain. The other two suspects were arrested in Romania and the UK.

0:50The Kilsec group conducted more than 1 ,000 cyber attacks since it began operating in 2024. The group exploited vulnerabilities, stole data, deployed ransomware and extorted victims. In other news, South Africa's air traffic control agency thwarted a ransomware attack against its OT network last week. The targeted network provides weather information to air traffic controllers. The Air Traffic and Navigation Services says the attack could have crippled both domestic and international air travel. The agency manages 6 % of the world's airspace. Authorities are investigating the possibility that an insider collaborated with the hackers.

1:35Misaligned open AI agents probed Canadian government websites but didn't breach any servers or internal systems, according to Canada's Cybersecurity Agency. The agency investigated reports of malicious AI agent activity after a similar incident in Australia. According to AI research lab Transluse, the probes targeted Library and Archives Canada, a Canadian federal agency. OpenAI has also notified dozens of organisations across the world that its AI agents probed their websites and might have exploited security flaws to access data. Meantime, OpenAI ignored warnings from its own employees earlier this year that new AI models were not being properly monitored during testing, according to a report from the New York Times.

2:26The report also claims OpenAI ignored warnings from external security researchers about its insecure infrastructure. American AI company Anthropix says the Chinese AI model Shipu GLM 5.3 has cyber capabilities similar to Mythos but ships with weak guardrails. According to Anthropic, threat actors can bypass GLM safety protocols and abuse the Chinese model to create exploits and launch cyber attacks. The company is urging governments to take action against open-weight models like GLM 5.3 and make sure they ship proper guardrails. Google has released Gemini 4 Argon, its newest frontier model. The company claims the model is highly capable at cyber defence tasks and is rolling it out to trusted cyber defenders through Google's Fairwind program.

3:20Google says Argon can autonomously find, validate and patch critical software vulnerabilities. Amnesty International has published a major report into spyware abuse by Morocco's domestic intelligence agency, DGST. Amnesty says the agency used a variety of methods to deploy NSO groups, Pegasus spyware, to the devices of human rights activists between 2017 and 2021. Morocco's state-owned telco, Morocco Telcom, was used in some cases to deploy Pegasus directly to targeted devices. The DGST operations were exposed by one of the agency's former employees. Hackers have leaked internal documents from Russian Cyber Research and Development Centre, Spetsvuz Avtomatika.

4:10The leaked files expose multiple projects to support Russian hacking operations. They include tools for target discovery, scanning, enrichment, active testing, internal network access and credential and data theft. The US Treasury sanctions Spetsvuz Avtomatica in May 2021 over developing hacking tools for Russia's SVR intelligence service. European law enforcement agencies are dropping contracts with Oxygen Forensics. The company's CEO is facing charges in the US over hiding Oxygen's Russian ownership when applying for government contracts. Romania's Anti-Corruption Agency, Latvia's State Police and London's Metropolitan Police are reviewing or are already in the process of terminating contracts with the company.

5:00Oxygen's products are also used in Germany, Spain, Italy, Poland and Hungary. A Dutch court has sentenced a 25-year-old man to a year and a half in prison for selling stolen credentials. The man sold more than 200 credential lists, containing a total of more than 1 million usernames and passwords. He also sold Grand Theft Auto 5 cheats and tools designed to crash GTA service and evade bans. The US has sanctioned 10 leaders of the trendy Aragua Venezuelan cartel for jackpotting ATMs. The group stole more than$40 million in 1 ,500 attacks. The sanctions come after the FBI arrested and charged more than 100 people in the past year accused of participating in the hacks.

5:48The gang members targeted isolated and unsupervised ATMs, connected laptops and infected them with the so-called Plotus malware. German authorities have suspended more than 9 ,300 phone numbers that were used in crypto investment scams. Officials have also taken regulatory measures against the relevant telcos. In total, authorities suspended almost 14 ,000 phone numbers used by scammers since December last year. Security researchers have found more than 13 ,000 screenshots taken by AI agents in public GitHub repositories. According to Glow Security, the agents were trying to help developers compare before and after screenshots of their apps.

6:33Agents installed in controlled or secure environments uploaded the images to GitHub to get around tight controls. inadvertently exposing sensitive corporate secrets. The affected companies include a Fortune 500 travel firm and a Frontier AI lab. Vulnerability disclosure and exploitation have both doubled in the last year, according to Google. The company says the increase is due to the integration of AI tools into the vulnerability discovery process. Google says AI is reliably finding more vulnerabilities, including ones that lead to remote code execution. A Chinese cyber espionage group has carried out an extensive spear phishing operation against AI experts working for US think tanks, universities and law firms.

7:22The group, known as TA419, impersonated economists and AI policymakers when approaching its targets. Proofpoint says this is the second Chinese APT it's recently seen targeting America's AI sector. Montenegro's government has approved the extradition of an Iranian hacker to the US. Amir Bharati was arrested in June in the Adriatic resort town of Koutour. US officials claim he's a member of Silent Librarian, an Iranian APT that hacks universities to steal research. Sesco has released patches for an actively exploited zero-day in Catalyst SD-WAN appliances. The vulnerability allows attackers to bypass API authentication via maliciously crafted HTTP requests.

8:12A successful exploit grants attackers admin-level access to the device's API management component. Hackers exploited two zero days in the Xamad helpdesk and issue tracking platform to breach Dutch cybersecurity non-profit DIVD. The vulnerabilities allow attackers to hijack sessions, elevate privileges and run remote code on Xamad servers. DIVD has reported the issues to the vendor, who's now working on patches. DIVD disclosed the hack over the weekend and said it suspects the attacker used AI tooling due to the speed at which the attack took place. And finally, internet infrastructure company Cloudflare is launching a certificate authority.

8:57The new CA will provide free certificates with automated issuance and renewal like Let's Encrypt. It'll also support quantum resistant cryptography. And that is all for this podcast edition. Today's show was brought to you by our sponsor, Portswigger. Find them at portswigger.net. And Risky Bulletin will be on a break next week. We'll return Monday, October 12. Bye for now.

9:28Bye.