AI-Powered Malware Scores Victims as Iran Hits US Critical Infrastructure

The day's security picture is shaped by AI operating on both sides of the line: a new remote access trojan now uses machine-learning scoring to triage its own victims, while a US government advisory warns that Iranian-linked hackers are actively disrupting domestic water and energy infrastructure. Running underneath both stories is an increasingly uncomfortable asymmetry — the same AI capabilities are accelerating attacker operations while guardrails constrain the researchers defending against them.

Security

A newly documented remote access trojan called Dolphin X ships with an AI-powered profiling module that scores and ranks infected hosts — automatically sorting victims by apparent value so operators can direct hands-on attention where it pays off. Automated victim triage at this layer has historically been manual or rudimentary; baking scoring into the malware itself is a meaningful operational efficiency gain for criminal networks, and a signal of what "AI-native" threat tooling actually looks like in practice. Notably, the adversarial developers building it face no content policy friction from the AI providers they may be using.

That adversarial pressure extends to physical infrastructure. An updated US government advisory warns that Iranian-linked threat actors are actively exploiting control systems at American water and energy providers — not probing, but disrupting. OT and ICS attacks with physical-consequence potential have been a documented risk for years, but active disruption against domestic utilities marks another step in the normalization of that threat category. Sector-wide, Australian energy provider Origin confirmed that an unauthorized party accessed and leaked customer PII — whether opportunistic or coordinated with the broader energy-sector targeting pattern isn't established, but the sector is under pressure from multiple directions simultaneously.

Two smaller items worth flagging: a Hacker News thread documents how Namecheap transferred a user's account to an unverified third party who simply asked — no identity verification, no MFA challenge. For anyone with domains registered there, this social-engineering surface deserves scrutiny. Separately, Microsoft responded to reports that certain LG monitors, when connected to a Windows PC, trigger a Windows Update installation of McAfee software through a vendor driver package — a reminder that the Windows Update delivery surface extends well beyond Microsoft's own patches.

AI

TechCrunch's reporting on AI guardrails and offensive security research should be read alongside the Dolphin X story. Security researchers who build exploit chains, analyze novel malware, and validate attack surfaces are finding OpenAI and Anthropic's content policies increasingly obstructive — even in clearly authorized pentesting contexts or active CVE research. The asymmetry is not theoretical: criminal operators deploying AI to score and rank their victims face no policy obstacles, while defenders navigate support escalations and policy exceptions they often don't receive. AegisAI's $36M Series A — raised by former Google security executives to counter AI-powered spear phishing — illustrates that the defense side recognizes the gap and is mobilizing capital to close it.

OpenAI launched ChatGPT Health to all US users, enabling integration with medical records and health-tracking platforms including Apple Health and MyFitnessPal. The company is making expansive clinical utility claims in its launch materials. The ambition is real — personalized health context makes LLM assistance meaningfully more relevant for care navigation — but LLM hallucination in high-stakes clinical contexts remains an unsolved problem that carefully worded announcements don't address.

Anthropic upgraded Claude voice mode to run on Opus and Sonnet rather than Haiku alone, and is extending voice access to third-party apps including Gmail and Slack. The previous Haiku-only restriction kept voice in the "fast but limited" bucket; running voice on frontier models is a different proposition. On the infrastructure side, Stripe is reportedly in acquisition talks with OpenRouter, the model-routing startup that lets developers address multiple AI providers through a single API — a move that would position Stripe as both the payment and routing layer for a large slice of AI applications.

Amazon shut down an internal AI agent research lab in a round of AGI-adjacent layoffs. The move reflects a broader rationalization underway at hyperscalers: the "fund every AI research thread simultaneously" phase is giving way to more deliberate portfolio management.

Tech

Google launched selfie video sign-in for account recovery, AI Avatar creation, and age verification. A short video rather than a static image is designed to raise the bar against spoofing. The feature is opt-in, but the direction is clear — Google is systematically migrating fallback authentication toward biometrics and away from SMS codes and backup key sets, which have documented weaknesses at scale.

Patreon laid off roughly 93 employees — approximately 20% of its workforce. CEO Jack Conte's memo describes the core business as strong and explicitly denies that AI is replacing human roles; the cuts are framed as a cost-structure adjustment to respond to market changes. Both things can be true simultaneously: a healthy core business and an AI-driven shift in content creation economics that changes the growth assumptions the headcount was built on.

Musk used Tesla's earnings call to outline robotaxi expansion — new cities, more miles, more unsupervised rides. The Verge's review of the actual deployment numbers is more grounding: Tesla's unsupervised mileage remains a fraction of Waymo's, and the geographic footprint is still narrow. The distance between earnings-call framing and operational deployment continues to widen.

Intel shares jumped 12% after Q2 results showed $4.5 billion in cash generated — a real positive data point in a turnaround narrative that has had plenty of setbacks. One strong quarter against sustained competitive pressure from AMD and TSMC-aligned rivals is progress, not a resolution.

Meta dropped out of a major clean energy industry coalition as its natural gas infrastructure investments accelerate. The hyperscale AI compute buildout is driving energy decisions across the industry that are increasingly incompatible with earlier decarbonization commitments — Meta is among the most explicit about the tradeoff, but it isn't alone.

The day's throughline is AI as force multiplier — for attackers scoring victims, for assistants reaching into medical records and voice interfaces, and for the energy consumption curves that are quietly rewriting the industry's own climate pledges.

Also yesterday

• Fake Job Interview Take-Home Projects Are Delivering Git Hook Malware
• Check Point SmartConsole Zero-Day Under Active Exploitation — Patch Now
• Nine-Year-Old Linux Kernel Flaw Grants Root on Default RHEL Installs — Patch Now
• Russian State Hackers Exploiting Patched Zimbra Flaw to Steal Email
• Russian State Hackers Exploited Zimbra Zero-Day to Steal Email and 2FA Codes