This week’s Department of Know is hosted by Rich Stroffolino, with guests Robb Dunewood, host, Daily Tech News Show, and David Cross, CISO, Atlassian.
Missed the live show? Check it out on YouTube.
The Department of Know is live every Friday at 4:00 p.m. ET. Join us each week by registering for the open discussion at CISOSeries.com.
In this week’s cybersecurity news…
OAuth tokens vulnerable on GitHub Dev
Security researcher Ammar Askar released details on a vulnerability that allows threat actors to install malicious VS Code Extensions that can steal GitHub OAuth tokens when using the GitHub.dev web-based source code editor. This is possible because the web editor does not scope tokens to a particular repository you’re interacting with. Threat actors can exploit a message-passing mechanism between VS Code windows and webviews using malicious JavaScript to extract the OAuth token. Microsoft was notified of the vulnerability on June 2nd, an hour before Askar released a proof of concept. He cited previous bad experiences with the Microsoft Security Response Center for the comically short notice.
The startup costs for US cyber force
According to a new report from the Commission on U.S. Cyber Force Generation as part of the Center for Strategic and International Studies, the cost of starting up a US Cyberforce is estimated to be around $11 billion. The report estimates that starting up the force would take 12 to 18 months and require at least 11,000 in staff The Government Accountability Office reported last year that roughly 61,000 people are currently involved in cyberspace operations across the Department of Defense, with a 2027 proposed budget of $7.7 billion.
In related news Homeland Security Secretary Markwayne Mullin said on Wednesday he would revitalize the agency, which has lost roughly one-third of its workforce and seen its $3 billion budget slashed during the current administration. The fiscal 2027 budget would cut more than $700 million from the agency. Mullin said that CISA probably needs “somewhere around” 2,800 employees.
ChatGPT share links used to host fake outage pages to deliver malware
Cybercriminals are abusing ChatGPT’s sharing feature to distribute malware through a campaign researchers call “LLMShare.” Attackers create legitimate-looking ChatGPT share pages hosted on OpenAI’s domain and use them to display fake service outage messages. Victims are told they can continue using ChatGPT by downloading a supposed desktop application, which is actually malware. Researchers found that the campaign leveraged online advertising and users’ trust in recognizable AI brands to increase success rates. This is a different ChatGPT hacking story from the Grey Vibe story presented on Friday.
HTTP/2 Bomb shows the wild world of AI vulnerability scanning
Security researchers at Calif discovered a denial-of-service technique dubbed “HTTP/2 Bomb” using OpenAI’s Codex. This attack impacts over 880,000 sites that use HTTP/2 and exploits a chain of three old 2016 vulnerabilities to launch compression-layer attacks and amplify rates. Unlike other classic “bomb” attacks, this attack floods a server with near-empty headers, exhausting the machine’s memory due to the per-entry bookkeeping it allocates. NGINX, Apache HTTPD, Microsoft IIS, Envoy, and Cloudflare Pingora are vulnerable to this attack, although at the time of this recording, NGINX, and Apache rolled out patches.
Huge thanks to our sponsor, Vanta
The good news: The Vanta Agent works like a GRC engineer in the background, finding every app your team uses, scoring the risk, and drafting fixes for you.
Vanta is the platform used by over sixteen thousand fast-moving companies like Ramp, Cursor, and Harvey who are shaping the future with AI, AND staying ahead of AI risk.
Get started at vanta.com/headlines.
Federal audit reveals NIST’s NVD problems
A report released on Thursday from the Department of Commerce found that NIST, the National Institute of Standards and Technology, has “mismanaged a critical cybersecurity vulnerability database through poor planning, inefficient operations, duplicate federal programs, and failure to communicate with users.” The National Vulnerability Database also known as NVD, “collects information about computer security flaws and adds details like severity ratings and affected products.” In February 2024, the database’s enrichment contract lapsed, creating a backlog of unprocessed security flaws that has only grown worse. NIST leaders admitted they had “no long-term plan for clearing the backlog, even as it grew from about 13,000 unprocessed security flaws in June 2024 to over 27,000 by the end of 2025.”
(Cyberscoop and Inspector General’s report)
Meta AI hands over Instagram account access
Attackers exploited Meta’s AI support chatbot to take over high-profile Instagram accounts, such as the Obama White House, the Space Force’s chief master sergeant, and Sephora. Attackers found that Meta used location as a verification factor for the chatbot. So they used a VPN to spoof a person’s real-world location, then just asked the bot to change the email address for the account and did a password reset.. Meta appears to have patched the flaw this week and has begun notifying affected users if they haven’t already.
(404Media)
Microsoft BUILD featured quantum and AI updates
Microsoft announced a new quantum computing chip called Majorana 2 at its BUILD conference. This chip uses lead-based materials for superconducting wires rather than aluminum, and claims to offer a 1,000-fold increase in certain performance metrics, compared to Majorana 1. Notably, Microsoft now estimates that production-capable quantum computers will be available sometime in 2029.
Microsoft also announced the Adaptive Spec Driven Scoring for Evaluation and Regression Testing, or ASSERT framework. This open-source framework is designed to make evaluating AI behavior easier, letting researchers use plain-language descriptions of expected behavior and policy, which are then turned into structured sets of behaviors, problem scenarios, and test cases, and then run against AI systems that produces a final score.
And finally, Microsoft also released the Agent Control Specification (ACS), which lets developers define policies for agents to follow in a single file. The idea is to create a common governance layer for agents, where admins can state what they can and cannot do, as well as when to bring a human into the loop.