In today’s cybersecurity news…
FBI warns of expanded Chinese hacking campaign
A joint advisory from Five Eyes allied security agencies, as well as the Czech Republic, Finland, the Netherlands, and Poland, warned that efforts by Salt Typhoon have expanded, hitting at least 200 organizations across 80 countries. These attacks were allegedly aided by three private companies that provided services to China’s People’s Liberation Army and the Ministry of State Security. Part of the reason there was a wide spread of victims is that these companies were allowed to choose their own targets. While Salt Typhoon notably gained long-term access to US telecoms, this advisory warned that the expanded attacks show a “broad, indiscriminate targeting of critical infrastructure.”
(WaPo)
AI-powered ransomware is a thing now
In “it was only a matter of time” news, researchers at ESET discovered a work-in-progress ransomware dubbed PromptLock. This is written in GoLang and uses the open-weight GPT-OSS:20b model from OpenAI. PromptLock utilizes hard-coded prompts to enable the model to dynamically generate Lua scripts. These scripts are then used for operations like file enumeration, inspection, exfiltration, and ultimately, encryption. The researchers cautioned that PromptLock, in its current form, isn’t entirely practical, requiring a local installation of Ollama with significant hardware resources to run and poor network segmentation to allow communication with an external server. Still mark your calendar, AI-powered ransomware will only get better after today!
Anthropic warns about “vibe-hacking”
The AI company released a new Threat Intelligence report, which warns that “Agentic AI systems are being weaponized.” The report profiled a threat actor using Claude Code to run a data extortion operation end-to-end, which targeted at least 17 organizations across various verticals within a month. Anthropic’s Claude chatbot was used for everything from technical consultation to crafting “psychologically targeted extortion demands.” The report also detailed the use of Claude by North Korean IT workers to get jobs at Fortune 500 US companies, and saw ads for romance scams using its chatbot on Telegram. While Anthropic created new controls to prevent similar types of abuse, it warned that the examples it found “likely reflect consistent patterns of behaviour across all frontier AI models.”
Swedish municipalities impacted by suspected ransomware
Miljödata (mil-yo-do-ta) is an HR software provider used by about 200 Swedish municipal governments. Its CEO, Erik Hallen, confirmed that threat actors are attempting to extort the company, seemingly as part of a ransomware attack. Several regional governments confirmed they use Miljödata to handle medical information and other sensitive employee data. Swedish Minister for Civil Defence Carl-Oskar Bohlin stressed that the scope of the incident is still under investigation, but said the government plans to present a new cybersecurity bill to parliament in the near future.
Huge thanks to our sponsor, Prophet Security
US DoD using software maintained by Russians
A new report from Hunted Labs found that the open-source tool fast-glob is solely maintained by a Yandex employee based in Russia. This helpful tool enables developers to perform actions on a group of files without requiring additional code. It’s a highly useful tool that the US Department of Defense utilizes in at least 30 pre-built software packages, as well as approximately 5,000 other projects globally, resulting in around 70 million downloads per week. Hunted Labs researchers found no malicious code in fast-glob and contacted the DoD’s Office of the Chief Information Officer three weeks before publishing findings. Over the summer, Secretary of Defense Pete Hegseth signed a memo directing DoD to “not procure any hardware or software susceptible to adversarial foreign influence.”
(NextGov)
Citrix RCE flaw under active exploitation
Citrix released updates for NetScaler ADC and Gateway devices to address a vulnerability that allows for remote code execution. The Shadowserver Foundation reports there are over 28,000 vulnerable devices online, with about 35% located in the US. Citrix did not provide any other mitigations, workarounds, or indicators of compromise. CISA and Citrix found evidence that these are already being exploited by malicious actors, and the flaw has already been added to the Known Exploited Vulnerabilities catalog. Federal agencies have until August 28th to patch.
Blind Eagle sinks its talons into Columbia
Researchers at Recorded Future published details about a campaign by the group Blind Eagle, which primarily targeted the Colombian government from May 2024 through July 2025. These attacks were carried out by five distinct clusters, using different infrastructure and operations, but with some overall common tactics based around using “cracked remote access trojans (RATs), dynamic domain providers, and legitimate internet services (LIS) for staging,” and using spearphishing lures posing as a government agency. Blind Eagle has been active since 2018, typically targeting victims in South America for financial gain.
When NDA stands for “New Download Attack”
Researchers at Check Point detailed a new campaign where threat actors deliver malware to American industrial and tech firms disguised as non-disclosure agreements. The threat actors initially approach victims through their “Contact Us” forms, posing as potential business partners and maintaining communication for several weeks. Eventually, they will ask the firm to sign an NDA, send it as a ZIP archive on Heroku, but containing a custom malware called MixShell. This appears to be a highly tailored approach; in some instances, the threat actors sent completely innocuous ZIP files, seemingly depending on the victim’s IP address or browser information. The threat actors set up fake websites using domains tied to real US businesses for added veracity.