<p><strong>TLDR:</strong> The US Commerce Department’s export control order forcing Anthropic to pull Fable 5 and Mythos 5 offline has become the defining story of the day, with 100+ security researchers signing an open letter calling the ban counterproductive and prediction markets giving 68-71% odds it gets reversed before July 1. The underlying trigger — a prompt-rephrasing technique to surface code vulnerabilities — is being widely characterized as not a true jailbreak, and Chinese competitor Zhipu AI’s stock surged 33% the day after the shutdown. Meanwhile, Cisco patched CVE-2026-20262, an actively exploited zero-day in Catalyst SD-WAN Manager that allows authenticated attackers to escalate to root via crafted file uploads. On the espionage front, Google’s GTIG exposed UNC6508, a China-linked actor that spent over a year inside North American medical and military research networks by backdooring REDCap servers and abusing Google Workspace compliance rules to silently BCC stolen emails to an attacker-controlled Gmail address.</p>
<hr>
<h2>Anthropic Fable 5 and Mythos 5 pulled under US export controls, 100+ researchers push back</h2>
<p>Three days after launch, the Commerce Department forced Anthropic to disable Fable 5 and Mythos 5 globally after Amazon researchers flagged a prompt-rephrasing technique that could surface code vulnerabilities. Security researchers — including Katie Moussouris — reviewed the underlying research and concluded the bypass was trivial and not meaningfully different from what other frontier models can do. Over 100 cybersecurity professionals signed an open letter arguing the ban disarms defenders while adversaries simply switch to open-source or foreign alternatives. The FreeBSD Foundation’s $250,000 AI-assisted vulnerability discovery project, which relied on Mythos access via Project Glasswing, is now directly affected. <a href="https://app.daily.dev/feed-by-ids?id=FmZW9X8yw&amp;id=38N9myGGx&amp;id=9hgCpPXLZ&amp;id=qUZGFtovj&amp;id=yVHu2auC9&amp;id=QdA56Uata&amp;id=lWmVNdGbs&amp;id=OL9Nt4CeM&amp;id=XVLyrF5fw&amp;id=U9qHMwXeK" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Cisco Catalyst SD-WAN Manager zero-day (CVE-2026-20262) actively exploited for root access</h2>
<p>Cisco patched CVE-2026-20262, a zero-day in Catalyst SD-WAN Manager being actively exploited in the wild — the second such zero-day in the product this month. The flaw stems from insufficient validation of user-supplied input during file uploads, letting authenticated remote attackers create or overwrite arbitrary files via crafted HTTP requests to an affected API endpoint, which escalates to root. All deployment types are affected: on-prem, cloud, and FedRAMP. Cisco PSIRT shared IOCs and advises checking logs for suspicious .jsp and .war file uploads. <a href="https://app.daily.dev/feed-by-ids?id=hlV1nkD7m&amp;id=HAAubkUIC" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>UNC6508 spent a year inside US medical and military networks, abused Google Workspace compliance rules for covert exfiltration</h2>
<p>Google’s GTIG and Mandiant disrupted a China-linked espionage campaign by UNC6508 that ran undetected from September 2023 to November 2025 across North American academic, medical, and military research institutions. The group backdoored externally facing REDCap servers with custom malware called INFINITERED, which persisted through software upgrades and harvested credentials. After reaching domain admin, they abused a legitimate Google Workspace content compliance feature to silently BCC matching emails to an attacker-controlled Gmail address — a technique that left virtually no forensic trace on the mail system. Defenders should patch REDCap, audit Workspace mail routing rules and admin logs, hunt for INFINITERED IOCs, and enforce phishing-resistant MFA on admin accounts. <a href="https://app.daily.dev/feed-by-ids?id=oTaLAXFBi&amp;id=hVASfIPrT&amp;id=bBP2HfWeM&amp;id=QLr5Au8iu&amp;id=qXRe9xEj7" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Microsoft 365 Copilot SearchLeak (CVE-2026-42824) enabled one-click data exfiltration via Bing SSRF</h2>
<p>Varonis disclosed SearchLeak, a three-stage attack chain in Microsoft 365 Copilot Enterprise Search that let an attacker exfiltrate emails, MFA tokens, calendar data, and SharePoint files with a single victim click. The chain combines a parameter-to-prompt injection via the search query URL, an HTML sanitizer race condition that lets an injected image tag fire before sanitization completes, and a server-side request forgery through Bing’s image search endpoint to bypass Content Security Policy. Microsoft patched it server-side — no customer action required — and assigned it CVE-2026-42824 with a critical severity rating. This is the second similar Varonis disclosure against Copilot, following the Reprompt attack, and mirrors the zero-click EchoLeak vulnerability. <a href="https://app.daily.dev/feed-by-ids?id=wRsuwvRWG&amp;id=L731prZby&amp;id=FfQgDzKcX" target="_blank" rel="noopener nofollow">Read more</a></p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>GlassWASM malware uses Solana blockchain as C2 dead-drop in trojanized Open VSX extensions:</strong> Socket’s Threat Research team found two trojanized VS Code fork extensions on Open VSX containing a TinyGo-compiled WebAssembly payload that polls the Solana blockchain’s public RPC API for attacker instructions embedded in SPL Memo fields — a blockchain-based C2 with no server to seize — then executes fileless download-and-execute commands via Node’s child_process (curl | bash on Unix, irm | iex on Windows). <a href="https://app.daily.dev/posts/kjdlX357b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AMD silently disabled TSME memory encryption on consumer Ryzen CPUs via AGESA firmware update:</strong> AMD removed Transparent Secure Memory Encryption from consumer Ryzen CPUs via a routine AGESA firmware update with no release notes, confirmed by MSI testing that the silicon is still capable but an internal flag forces it off — the BIOS toggle now does nothing on consumer hardware, and AMD has only stated TSME is now a PRO-only feature without explaining the silent removal. <a href="https://app.daily.dev/feed-by-ids?id=vizheGgV4&amp;id=6Ni6ARQuG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ShinyHunters claims 429,000+ Council of Europe documents including payslips and bank details:</strong> ShinyHunters threatened to leak allegedly stolen Council of Europe data on June 16, 2026, claiming 409,000+ payslips for 10,000+ staff spanning 2011-2026, 3,700+ personnel files, and sensitive financial and medical records — the Council of Europe has confirmed it is investigating. <a href="https://app.daily.dev/posts/psijygyZr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Langflow CVE-2026-5027 under active exploitation by MuddyWater, ~7,000 instances exposed:</strong> A high-severity path traversal flaw in Langflow’s file upload endpoint is being actively exploited by Iranian state-sponsored group MuddyWater despite a patch shipping over two months ago — exploitation is trivial because Langflow ships with auto-login enabled by default, and roughly 7,000 internet-exposed instances remain unpatched; upgrade to version 1.9.0 or later. <a href="https://app.daily.dev/posts/jzjaNnQnB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SimpleHelp CVE-2026-48558 allows unauthenticated creation of privileged technician accounts on ~14,000 exposed servers:</strong> A critical flaw in SimpleHelp versions 5.5.15 and older lets unauthenticated attackers create privileged technician accounts on servers using OIDC authentication, bypassing MFA entirely; patches are available as of June 9 (versions 5.5.16 and 6.0RC2), and roughly 7.2% of the ~14,000 internet-exposed SimpleHelp servers use OIDC. <a href="https://app.daily.dev/posts/EUpUXxqvQ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>


CVE disclosures, supply-chain attacks, security incidents, and developer-facing security tooling.

Security Digest

A security-focused weekly roundup covering several major stories: the US Commerce Department forced Anthropic to pull Fable 5 and Mythos 5 offline over a prompt-rephrasing technique flagged as a security risk, drawing pushback from 100+ researchers who call the ban counterproductive. Cisco patched CVE-2026-20262, an actively exploited zero-day in Catalyst SD-WAN Manager enabling root escalation via crafted file uploads. Google's GTIG exposed UNC6508, a China-linked group that spent over a year inside North American medical and military networks by backdooring REDCap servers and abusing Google Workspace compliance rules for covert email exfiltration. Additional items cover a Microsoft 365 Copilot SSRF chain (CVE-2026-42824), blockchain-based C2 malware in VS Code extensions, AMD silently disabling memory encryption on consumer Ryzen CPUs, a ShinyHunters data breach claim against the Council of Europe, active exploitation of Langflow and SimpleHelp vulnerabilities.

Anthropic Fable 5 ban dominates the week, Cisco SD-WAN zero-day under active attack