This week’s TL;DR covers two stories making waves in the security community. First, Anthropic’s release of Claude Fable 5 and Mythos 5, and what security experts are actually saying about it. Second, a supply chain attack targeting over a million WordPress sites through tampered plugin scripts.
Claude Fable 5 Doesn’t Change the Security Story
Anthropic this week released Claude Fable 5 and Mythos 5, the latest iterations of its frontier AI models. Mythos 5 is a direct upgrade on Mythos Preview: the model Anthropic had quietly rolled out in April under Project Glasswing, restricting access to a small pool of trusted partners, including the US government, due to its ability to identify and exploit vulnerabilities in popular software. Fable 5 is essentially the same model but with additional safeguards, described by Anthropic as Mythos “made safe for general use.”
Those safeguards include a new set of safety classifiers; separate AI systems designed to intercept misuse before Fable 5 can generate a harmful output. When users query certain cybersecurity topics, the model may silently fall back to the previous Claude Opus 4.8. Anthropic says this triggers in under 5% of cases and users are notified when it happens.
The company also claims that after more than 1,000 hours of internal and external red teaming, penetration testers were unable to produce any universal jailbreaks, though they acknowledge the UK’s AI Security Institute made progress toward one in initial testing.
Experts largely agree this doesn’t fundamentally change the threat landscape. Rob T. Lee, chief AI officer at SANS Institute, operates under the assumption that Mythos-caliber models have already made it into the wrong hands. Rich Mogull of the Cloud Security Alliance put it plainly: “The release of Fable did not make you less secure than you were the day before.”
The advice from the CSA remains the same as it was in April. That is, focus on security basics: segmentation, egress filtering, multifactor authentication, and defense in depth. AI won’t render those irrelevant; it will make them more important.
Claude Fable 5 – TL;DR
Anthropic has released Claude Fable 5 and Mythos 5, with Fable being the safety-gated public version of its most capable model yet. Despite concerns about AI-assisted cyberattacks, security experts say the threat landscape hasn’t materially changed overnight. The guidance remains the same: patch fast, layer your defenses, and assume capable models are already in adversarial hands.
Popular WordPress Plugins Backdoored in Supply Chain Attack
A supply chain attack targeting three widely-used WordPress plugins, PushEngage, OptinMonster, and TrustPulse, was disclosed by security firm Sansec on June 13. All three are operated by the same company, Awesome Motive. Together they reach more than 1.2 million WordPress sites, with OptinMonster alone accounting for over a million active installs.
The attacker tampered with JavaScript files served via CDN to those plugin customers. The malicious code was designed to activate only when a logged-in WordPress administrator loaded the page. At which point it used that admin’s active session to create a new admin account under the attacker’s control, install a hidden plugin not visible in the WordPress dashboard, and exfiltrate the new credentials to a fake domain (tidio[.]cc) registered weeks before the attack. The hidden plugin opens a web shell, giving the attacker persistent remote code execution on any compromised server.
The exposure window varied. OptinMonster and TrustPulse were affected for roughly 25 minutes on June 12. PushEngage had a longer window of several hours on June 12, and some CDN servers were still serving the tampered script into June 14.
PushEngage attributed the initial breach to an attacker exploiting a vulnerability in UpdraftPlus (CVE-2026-10795, rated 8.1 high) on its marketing website, from which they were able to steal a CDN API key and modify the files being delivered to customer sites. Sansec considers the entry point still unconfirmed. OptinMonster and TrustPulse had issued no official guidance as of publication.
Any site running these plugins during that window should be treated as potentially compromised. Because the backdoor is designed to hide from the WordPress dashboard, the only reliable check is a direct server-side scan.
WordPress Plugin Attack – TL;DR
Attackers tampered with CDN-served JavaScript. In total, for three major WordPress plugins, backdooring sites whose admins happened to be logged in during the attack window. With over 1.2 million sites potentially exposed and a hidden web shell as the payload, any affected site should conduct a server-side forensic check.
Conclusion
These two stories reflect the pressures facing security teams right now. AI is expanding the attack surface on one side, and supply chain compromises targeting trusted software on the other. Neither is entirely new, but both are accelerating rapidly. Endpoints and the software running on them remain among the most vulnerable points in any organization’s security stack. Get in touch to learn how ThinScale can help you keep them under control.
