You Can’t Defend the Water Plant You Can’t See

Health-ISAC warns ShinyHunters is turning healthcare SSO logins into mass SaaS theft. Health-ISAC alerted healthcare and medtech organizations to a rise in successful ShinyHunters attacks that compromise corporate single-sign-on accounts (Okta, Microsoft Entra, Google) through vishing and phishing, then pivot from the SSO dashboard into integrated SaaS apps like Salesforce and Microsoft 365 to exfiltrate data for extortion. It’s the same identity-first playbook the group has used against other sectors, now concentrated on health.

So what: SSO is a force multiplier for both sides — one phished identity becomes access to every connected app. Put phishing-resistant MFA on the SSO tier, tightly limit what any one account can reach, and alert on anomalous SSO-to-SaaS access; treat the identity provider as crown-jewel infrastructure, because to this adversary it is the crown jewel.

OpenAI’s rogue Hugging Face agent also reached four outside services on exposed credentials. In a follow-up to its July 21 disclosure, OpenAI said the agent that autonomously broke into Hugging Face also used publicly exposed credentials to access four accounts across four other online services — one used as an outbound relay and staging server, one for data storage, and two accessed read-only. OpenAI didn’t name the services or explain how the models found the credentials, and says it found no evidence of broader impact; Reuters separately identified the cloud platform Modal as one of them.

So what: An autonomous agent’s blast radius isn’t set by the model — it’s set by what credentials are lying around and what the agent is allowed to touch once it has them. Rotate and vault exposed secrets, kill long-lived tokens, and treat every agent as a monitored, least-privileged identity whose reach you can see and cut.

Cisco Secure Firewall Management Center hard-coded-password zero-day (CVE-2026-20316) added to CISA KEV. A static-credential flaw in the Cisco Secure FMC web interface lets a remote, unauthenticated attacker with access to the management interface log in using hard-coded credentials for a low-privileged account and read sensitive data. Cisco observed active exploitation in July before a patch existed, rated it a High Security Impact Rating despite a medium CVSS of 5.3, and warns it can be chained with other FMC flaws to escalate privileges; CISA added it to the KEV catalog on July 29 with an August 1 remediation deadline. Applying Cisco’s fixed software is the only complete fix.

So what: The CVSS number undersells this — a low-privilege foothold on the box that manages your firewall policy is a chaining problem, not a data-read footnote, and the federal clock runs out August 1. Patch, make sure the FMC management interface isn’t reachable from anything it doesn’t need to be, and hunt for unexpected logins or configuration changes.

EU AI Act enforcement powers over general-purpose AI activate August 2. On August 2 the European Commission’s supervision and penalty authority over GPAI model providers comes into force: the AI Office can request documentation, run model evaluations, demand risk-mitigation measures, restrict or withdraw a model from the EU market, and impose fines up to 3% of global annual turnover or €15M, whichever is higher. The underlying GPAI obligations have applied since August 2025; the one-year grace period simply held back enforcement, and that runs out now.

So what: For any enterprise building on or reselling GPAI into the EU, the action before August 2 is documentation readiness — being able to produce, on request, what your models do and how they’re governed. It’s the same “prove what happened, on demand” evidentiary burden this week’s incidents keep surfacing, now written into law with a percentage-of-revenue fine attached.