Threat Brief — 2026-08-19 — AI Weapons Hit OT and Cloud

U.S. agencies are warning that AI-generated exploit scripts are actively targeting Siemens PLCs in critical infrastructure — a significant escalation in AI-assisted operational-technology attacks. A mass compromise of 14,500+ Dahua IP cameras underscores how legacy IoT remains a persistent attack-surface problem. On the cloud side, CISA formally added an MLflow SSRF flaw to its Known Exploited Vulnerabilities catalog, and a remote Spectre attack against Cloudflare Workers demonstrates that shared-tenant isolation remains brittle.

Top items

• CISA warns of AI-powered attacks on Siemens S7 PLCs in U.S. critical infrastructure. Threat actors are using AI-generated scripts to exploit Siemens S7 Series programmable logic controllers, marking a notable convergence of AI tooling and OT targeting. U.S. cybersecurity agencies issued the advisory; affected products include Siemens S7 Series PLCs deployed in critical infrastructure. (src: BleepingComputer)

• CISA adds MLflow SSRF (CVE-2026-64849KEV) to Known Exploited Vulnerabilities catalog. This is a formal KEV addition for the MLflow server-side request forgery vulnerability that was first reported as actively exploited on 2026-08-18 by The Hacker News. The KEV listing confirms in-the-wild exploitation and triggers federal patching deadlines. [CVE-2026-64849KEV in CISA KEV] (src: CISA)

• 14,500+ Dahua IP cameras compromised in 35-day "CameraSwarm" campaign. Hackers mass-compromised Dahua web cameras — predominantly in Ukraine and Russia — over a 35-day window, exposing surveillance feeds and logs. The scale highlights continued exploitation of unpatched IoT firmware in conflict-zone adjacent regions. (src: BleepingComputer)

• CareCloud data breach impacts 3.7 million patients. U.S. healthcare IT company CareCloud disclosed that a breach suffered earlier this year exposed personal and medical data of 3.7 million individuals — one of the larger healthcare breaches this year. Affected entity: CareCloud (healthtech). (src: BleepingComputer)

• Sakura Internet breach exposes up to 1.36 million accounts. Japanese cloud and data-center provider Sakura Internet disclosed that attackers accessed its sales management system containing customer contract and membership data. The breach raises supply-chain concerns for organizations hosting on Sakura's infrastructure. (src: BleepingComputer)

• Remote Spectre attack against Cloudflare Workers leaks JWT at 12 bits/sec. Researchers demonstrated a remote Spectre side-channel attack extracting a JWT from a co-located Cloudflare Worker in production at up to 12 bits/second, 360 times slower than local attacks but still practical. This challenges assumptions about isolation in serverless multi-tenant environments. (src: The Hacker News)

• OpenAI pauses frontier reinforcement-learning training to shore up AI safety defenses. OpenAI disclosed a two-week pause in RL training for its latest models while it expanded monitoring and defensive guardrails to prevent unsafe AI behavior. This signals ongoing tension between model capability and safety controls at frontier labs. (src: The Hacker News)

• "Kriminal" AI platform offers guardrail-free cybercrime tooling for cryptocurrency. A new AI service provides unfiltered social engineering, offensive cybercrime, and OSINT scanning capabilities to anyone willing to pay in crypto, despite nominal terms forbidding illicit use. This lowers the barrier to entry for AI-assisted attacks. (src: Dark Reading)

• Rogue ransomware affiliate "Ransom Busters" poses as data-recovery firm. A suspected ransomware affiliate contacts victims before attacks go public, offering decryption keys and data deletion for $20K–$60K — but delivers neither. This was first reported 2026-08-18 by Dark Reading; today's BleepingComputer coverage adds further victim-detail corroboration. (src: BleepingComputer)

Themes

AI as both weapon and target. Three of today's items center on AI: "Kriminal" weaponizes AI for cybercrime-as-a-service, AI-generated scripts are being used to attack Siemens PLCs, and OpenAI itself paused RL training over safety concerns. The dual-use pressure on frontier models is intensifying on both offensive and defensive fronts.

Shared infrastructure isolation under pressure. The Cloudflare Workers Spectre attack and the MLflow SSRF KEV addition both illustrate that multi-tenant and shared-platform assumptions remain exploitable — whether through microarchitectural side channels or server-side request forgery.