Three Security Threats Converge as Apple Sues OpenAI Over Trade Secrets

Three independent security threats — a DoS flaw baked into every unpatched OpenSSL server, a blockchain-backed supply chain attack on the Vite npm ecosystem, and a botnet specifically hunting exposed AI inference endpoints — arrived simultaneously yesterday, while on the business side Apple filed a trade secrets lawsuit that could cast a shadow over OpenAI's IPO path. The day's pattern: the attack surface is expanding faster than defenders are closing it.

Security

OpenSSL HollowByte is the patch-now story. A malformed TLS ClientHello message of just 11 bytes causes vulnerable OpenSSL servers to allocate up to 131 KB of memory for a TLS record that never arrives. On glibc-based Linux systems — the majority of server deployments — that memory is not reclaimed until the process restarts. The math is stark: an attacker who can send enough 11-byte payloads can exhaust a server's memory without completing a single handshake, producing a denial-of-service condition at negligible bandwidth cost. OpenSSL shipped the fix in July; any deployment still on an unpatched build is exposed, and a process restart is required after patching — not just the package upgrade. (The Hacker News, BleepingComputer)

While HollowByte targets infrastructure universally, ViteVenom targets the JavaScript developer supply chain specifically. Checkmarx identified seven malicious npm packages impersonating Vite tooling. The distinguishing technical detail: command-and-control instructions are stored on the blockchain rather than a conventional server. That makes the C2 channel effectively takedown-resistant — you can pull the npm packages, but you cannot kill the instruction channel. The payload is a remote access trojan. Any developer who installed these packages and has not audited their environment should treat it as a compromise and act accordingly. (The Hacker News)

The third campaign, NadMesh, targets the AI services layer directly. The Go-written botnet continuously populates its scan queue via Shodan, looking for unauthenticated or weakly-protected instances of ComfyUI, Ollama, n8n, Open WebUI, and Langflow. When it finds them, it extracts any cloud credentials visible in the environment — AWS keys, Kubernetes tokens. The operator's own dashboard reportedly claims 3,811 unique AWS keys harvested. NadMesh is a direct consequence of the "run AI locally, share it with the team" pattern that routinely skips authentication steps. If your AI inference service has any external exposure, assume it has been scanned — and likely accessed. (The Hacker News)

Two supporting items round out the security picture. Abbott Laboratories disclosed unauthorized access to internal legacy Exact Sciences systems in its Cancer Diagnostics unit and is simultaneously investigating an extortion claim against a separate system — two active incidents at once, with details still limited. (BleepingComputer) And a researcher disclosed that TP-Link Kasa EC71 cameras have been transmitting their owner's home GPS coordinates over an unauthenticated UDP port for six years — plaintext, no authentication required, accessible to anyone who could reach the port. Six years is a long time for a home location to be leaking silently. (GitHub)

AI

The capital story is Databricks reaching a $188 billion valuation on a new $3 billion funding round — the largest private valuation in enterprise AI infrastructure right now. What distinguishes this number is that Databricks earned part of it by publishing credible open-weight model research arguing these models meaningfully cut coding costs, giving it a differentiated narrative at a time when many AI companies are competing on assertions alone. The market is making a specific bet: that the data and AI infrastructure layer, not the model layer itself, is where durable margin accumulates. (TechCrunch, The Information)

Isomorphic Labs — Alphabet's drug design company — published its Drug Design Engine, positioned as a generative step beyond AlphaFold's structure-prediction capabilities. The direction of travel is meaningful: from predicting what proteins look like to designing molecules that bind them. Whether this translates into clinical candidates is an open empirical question, but the computational approach is genuinely distinct from what preceded it. (Isomorphic Labs)

On the labor side, Kaiser Permanente nurses are reporting concrete friction with AI deployment in clinical settings — not abstract unease but specific claims that automated monitoring and AI workflow tools are degrading care quality and worsening working conditions. As AI integrates deeper into high-stakes environments, frontline signal from the people running the workflows matters more than adoption metrics. (Local News Matters)

Tech

Apple filed a trade secrets lawsuit against OpenAI last Friday, and the complaint is detailed and pointed. The allegations reach to OpenAI's chief hardware officer and describe what Apple characterizes as systematic misappropriation of confidential information. The legal theory relies on trade secrets rather than patent or copyright — harder to prove, but it generates significant discovery pressure and creates uncertainty for a company managing IPO preparations in parallel. Apple is also separately in settlement discussions with the DOJ over its antitrust case, making this an unusually active moment in Apple's legal posture. (TechCrunch, The Information)

The FAA restored Boeing's authority to sign off on its own airworthiness certificates for the 737 MAX and 787, reversing the post-accident enhanced oversight regime. The agency determined Boeing's quality management system has been sufficiently rehabilitated. Given the history of how that oversight tightening came about, this decision will be watched closely by anyone who flies either aircraft type. (CNBC)

FireSat launched — the Google-backed wildfire detection constellation designed to spot fires smaller than current satellite systems can resolve, with faster refresh cycles. The timing is not incidental: active wildfire smoke is affecting large portions of the US and Canada right now, and earlier detection at smaller fire size is the difference between a managed response and a runaway event. (Ars Technica) Also noted: Apple raised Apple Music prices — individual plans to $11.99/month (+$1), family plans to $19.99/month (+$3). (The Verge)

The attack surface expanded at four independent layers yesterday: TLS server infrastructure, the JavaScript package ecosystem, self-hosted AI services, and IoT cameras sitting in people's homes. Patch lists are longer than usual; work through yours.

Also yesterday

• CVE-2026-25089: FortiSandbox Unauthenticated RCE Added to CISA KEV — Patch Now
• CISA Flags Actively Exploited SharePoint RCE Zero-Day CVE-2026-58644 — Patch Now
• CISA Orders Immediate Patching of Actively Exploited Fortinet FortiSandbox Flaws