<p><strong>TLDR:</strong> Meta’s AI support chatbot was exploited to take over high-profile Instagram accounts — including Barack Obama’s White House account — by simply asking it to swap the linked email address, no technical skill required. A PyPI supply chain attack called Shai-Hulud’s Hades branch hid credential-stealing payloads in 37 wheel artifacts across 19 packages, executing at interpreter startup via .pth hooks before any import. Silent Ransom Group is running callback phishing campaigns against U.S. law firms, exfiltrating data and sending extortion demands within 30 minutes of access. The C0XMO botnet is actively exploiting a DD-WRT buffer overflow to build a multi-architecture DDoS fleet that also kills competing malware on infected hosts.</p>
<hr>
<h2>Meta AI chatbot exploited to hijack high-profile Instagram accounts</h2>
<p>Attackers took over Instagram accounts belonging to Barack Obama’s White House, the Chief Master Sergeant of Space Force, and Sephora by asking Meta’s AI support bot to change the email address on the target account. No technical exploitation was involved — just a support conversation and a username. Meta rolled out AI-powered account recovery to all Facebook and Instagram users in March, giving the bot authority to reset passwords and modify account credentials. Victims report no path to a human agent, which is the real problem: delegating irreversible account security operations to a chatbot with no escalation path is a design failure, not just a bug. <a href="https://app.daily.dev/posts/4aXIqPGNb" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Shai-Hulud Hades worm plants credential stealer in 19 PyPI packages via .pth startup hook</h2>
<p>The Hades branch of the Shai-Hulud worm spread across 37 wheel artifacts in 19 PyPI packages, including legitimate research tools like dynamo-release, spateo-release, coolbox, and ufish. The payload runs at Python interpreter startup through a .pth hook — a single pip install is enough, no import needed. Once running, it downloads the Bun runtime and executes an obfuscated JavaScript stealer targeting GitHub tokens, PyPI/npm credentials, AWS/GCP/Azure/Kubernetes secrets, SSH keys, and AI tool configs, exfiltrating everything to attacker-controlled public GitHub repos. Immediate actions: check for unexpected .pth entries, rotate secrets in the priority order the post outlines, and enforce hash-pinned lockfiles with sandboxed installs in CI. <a href="https://app.daily.dev/posts/coqLi3SH8" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Silent Ransom Group targets law firms with callback phishing and 30-minute extortion clock</h2>
<p>Silent Ransom Group (UNC3753 / Luna Moth) is running vishing campaigns against U.S. law firms, starting with invoice-themed phishing emails and following up with phone calls where attackers pose as IT help desk staff. Victims are walked into remote support sessions via Teams, Zoom, or Quick Assist, then convinced to install AnyDesk or Zoho Assist. From there the group exfiltrates legal and financial documents using WinSCP or Rclone and sends extortion demands within 30 minutes of leaving — with a three-day deadline. The fast-flux DNS infrastructure protecting their leak platforms makes takedown difficult. Mandiant and the FBI recommend strict IT verification procedures and hard limits on which remote access tools employees can install. <a href="https://app.daily.dev/posts/WTSTzyOXE" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>C0XMO botnet exploits CVE-2021-27137 in DD-WRT routers, kills rival malware on infected hosts</h2>
<p>A new Gafgyt variant called C0XMO is exploiting CVE-2021-27137, a buffer overflow in DD-WRT firmware, to spread across ARM, MIPS, x86, and other architectures. It supports 19 DDoS methods including UDP/TCP/SYN floods and NTP/Memcached amplification, and uses a Python-based scanner to brute-force SSH/Telnet for lateral movement. What makes it stand out: it actively hunts and terminates competing botnet clients and red-team tools on infected hosts, removes their persistence, and installs its own via cron and shell startup files. If you have DD-WRT devices in your environment, update firmware, rotate credentials, and disable remote access you’re not actively using. <a href="https://app.daily.dev/posts/h3AopVAei" target="_blank" rel="noopener nofollow">Read more</a></p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>OpenAI Lockdown Mode coverage: already in previous digest:</strong> This story was covered in the previous digest; the new post adds no material changes to the feature or its limitations. <a href="https://app.daily.dev/posts/79V4ztP0j" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>2026 breach roundup names supply chain attacks on Trivy and Bitwarden as cascading to OpenAI and Vercel:</strong> A mid-year breach summary documents a supply chain attack wave that compromised Trivy and Bitwarden and cascaded downstream to OpenAI and Vercel, alongside a Chinese breach of an FBI surveillance system and ShinyHunters hitting Canvas affecting 30 million students. <a href="https://app.daily.dev/posts/HUI7Hjrsk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code introduces OWASP Top 10 flaws in 45% of cases, per vibe coding governance piece:</strong> A governance writeup on AI-assisted development in journalism cites a 45% rate of OWASP Top 10 vulnerabilities in AI-generated codebases, recommending scope boundaries, periodic expert review, and documentation requirements for any team relying on AI to write production code. <a href="https://app.daily.dev/posts/iTHJ3nyYe" 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 digest covering four major incidents: Meta's AI support chatbot was exploited to hijack high-profile Instagram accounts (including Obama's White House account) by simply requesting email swaps with no technical skill; the Shai-Hulud Hades worm planted credential-stealing payloads in 37 wheel artifacts across 19 PyPI packages using .pth startup hooks to steal GitHub tokens, cloud secrets, and SSH keys; Silent Ransom Group is running callback phishing campaigns against U.S. law firms, exfiltrating data and sending extortion demands within 30 minutes; and the C0XMO botnet exploits a DD-WRT buffer overflow (CVE-2021-27137) to build a multi-architecture DDoS fleet that also kills competing malware on infected hosts.

Meta AI hands over Instagram accounts on request, Shai-Hulud worm hits PyPI