In today’s cybersecurity news…

A Patch Tuesday for the books

Microsoft released a record June Patch Tuesday with nearly 200 security fixes, including around 35 critical vulnerabilities and several flaws that already have public exploit code available. AI-assisted bug hunting is accelerating discovery, which could make bigger patch cycles more common. The update also addresses several exploits published by Nightmare Eclipse, who claims to be a former Microsoft employee and has promised to release more zero-days. Rapid7 researchers note the patch total was larger, with Microsoft separately fixing about 360 browser vulnerabilities that aren’t included in the official Patch Tuesday count. (Krebs on Security)

‘Nightmare Eclipse’ drops Windows 0-day

Yep, just hours after Microsoft’s big Patch Tuesday, ‘Nightmare Eclipse’ released exploit code for a new Microsoft Defender flaw they claim works on fully patched Windows 10 and 11 systems. The bug could let an attacker gain SYSTEM-level control of a machine, and early security researcher Will Dormann of Tharros Labs suggests the exploit is real, though not completely reliable. ‘Nightmare Eclipse’ backed away from earlier threats of a big July vulnerability dump, saying the latest exploit took more time and effort than expected. (The Register)

Claude Fable restricted at Microsoft

The Verge reports Microsoft is restricting employee access to Anthropic’s new Claude Fable 5 model because its safety system requires prompts and outputs to be retained for at least 30 days – that’s unlike earlier Claude models that support zero-data-retention policies. Microsoft has already made Fable 5 available to GitHub Copilot and Foundry customers, but its legal teams are still reviewing whether the model can be safely used internally, given concerns around confidential and customer data. Anthropic says Fable 5 needs additional safety controls, but those controls come with data retention requirements. (The Verge)

ServiceNow unpacks data security incident

ServiceNow briefly warned customers that a security issue may have been used to access data in some customer environments, but the company now says the activity was most likely caused by security researchers conducting bug bounty research rather than malicious attackers. The issue could allow unauthenticated access to certain information under specific conditions and was patched on June 5, affecting a limited number of customers. ServiceNow says multiple researchers independently reported the same flaw through bug bounty programs, and its investigation found no evidence that any data was retained or misused. (Dark Reading)

Huge thanks to our sponsor, Doppel

Langflow flaw exploited

Attackers are actively exploiting a high-severity flaw in Langflow, the popular open-source platform used to build AI apps, agents, and RAG workflows. Tenable, which discovered the flaw, says it lets attackers write arbitrary files to a server through a path traversal bug, and VulnCheck researcher Caitlin Condon says attackers are already dropping test files on vulnerable systems. Langflow enables unauthenticated auto-login by default, meaning attackers don’t need credentials to reach the vulnerable endpoint, and Condon says scans have identified roughly 7,000 publicly exposed instances. The bug has been patched for months, but it’s another example of attackers targeting AI development infrastructure as adoption grows. (BleepingComputer)

China ramps up AI espionage

CrowdStrike reports that China-linked hacking groups are targeting U.S. tech companies more and more to steal AI technology and intellectual property, with Chinese actors responsible for more than 58% of the state-sponsored cyberattacks it tracked against the sector over the past year. Crowdstrike says the goal is to help China close the AI gap with the U.S., especially as export controls continue to limit China’s access to advanced AI chips. Chinese groups are also reportedly targeting government communications in Southeast Asia and maintaining long-term access to North American tech companies by exploiting vulnerabilities. (CNBC)

CISA wants federal agencies to patch, lickity-split

The agency is tightening federal patching requirements, giving other agencies three days to fix the highest-risk vulnerabilities, specifically flaws that are actively exploited, internet-facing, automatable, and capable of giving attackers control of a system. CISA officials say this is necessary because AI is making it easier for attackers to find and exploit vulnerabilities at scale. CISA says not every bug needs urgent attention, but those dangerous ones can’t sit unpatched for weeks anymore. (The Record)

‘Miasma’ worm source code leaked on GitHub

The source code for Miasma, a sophisticated credential-stealing worm linked to supply-chain attacks against Red Hat npm packages and dozens of Microsoft GitHub repositories, was briefly published on GitHub through compromised developer accounts. SafeDep researchers say the release appears intentional and reveals a highly advanced framework that steals cloud and developer credentials, compromises software packages and repositories, and even targets coding tools like Claude, Gemini, Cursor, Copilot, and Cline. Much like the earlier Shai-Hulud leak that led to new variants and attacks, making Miasma public could do the same. (BleepingComputer)