Saturday 20 June 2026

usbliter8: The Unpatchable iPhone Exploit That Will Never Get a Patch

A working unpatchable BootROM exploit for millions of iPhones drops, Western Australia deploys real-time facial recognition, and Microsoft's AutoJack attack turns AI agents into remote code execution vectors.

Lead story

usbliter8: The Unpatchable iPhone Exploit That Will Never Get a Patch

Security researchers at Paradigm Shift have published a working exploit called usbliter8 that achieves arbitrary code execution inside the SecureROM of Apple's A12 and A13 chips — the ones inside the iPhone XS through iPhone 11 family. If that sounds familiar, it should: this is the same class of vulnerability as checkm8, the 2019 BootROM exploit that still haunts every A10-era device in service today.

The critical detail is in the word "burned." SecureROM code is written into the silicon at manufacture. Apple cannot update it remotely, push a fix, or issue a patch. Every affected device will carry this flaw until it's decommissioned. That's tens of millions of handsets globally — many still actively used by individuals, businesses, and government agencies.

What it actually takes to exploit it. This is not a remote attack. The exploit requires physical USB access to the device, which significantly limits casual mass exploitation. Think targeted scenarios: border searches, device seizure, physical theft, or a charging cable left somewhere interesting. The jailbreak community will also have a field day, which is a secondary consequence rather than a threat.

Why it still matters. "Physical access required" is not the same as "nothing to worry about." Law enforcement and intelligence agencies routinely use physical-access BootROM exploits for device forensics — tools like GrayKey are built on exactly this kind of vulnerability. The public release of a working exploit means that capability, previously in the hands of a small number of well-resourced actors, is now available to anyone with the technical inclination to use it.

For enterprises and government agencies, the calculus is straightforward: any A12/A13 device that contains sensitive data and could plausibly end up in an adversary's hands is now a higher-risk device than it was 48 hours ago. The Australian Signals Directorate's advice on device disposal and mobile device management policies becomes newly relevant here — especially for agencies running older iPhone fleets that haven't yet been refreshed.

What Apple can do. Not much, directly. Apple has reportedly acknowledged the research. Future hardware won't be affected — A14 and later chips don't share this BootROM code. The practical mitigations are organisational: enforce full-disk encryption (already default on iOS), ensure strong alphanumeric passcodes rather than six-digit PINs, enable Lockdown Mode on high-risk devices, and accelerate device refresh cycles for sensitive roles.

The broader pattern. BootROM exploits have a long tail. checkm8, disclosed in 2019, is still being actively used in forensic tools and jailbreaks today. usbliter8 will likely follow the same trajectory — quietly becoming infrastructure for a decade's worth of targeted attacks on devices that will never see a fix.

The lesson for anyone managing a device fleet: "end of software support" and "end of security risk" are not the same date, and they never have been.

Also today

Western Australia Deploys Australia's First Real-Time Facial Recognition Cameras

Western Australia Police will become the first law enforcement agency in Australia to operate real-time facial recognition surveillance cameras in public spaces, according to reporting from The Mandarin. Civil liberties groups have labelled the move a "massive overreach," arguing the technology cannot reliably distinguish between individuals at the accuracy thresholds needed to justify its use in identifying suspects in live settings. The deployment puts WA ahead of any federal framework governing the technology — Australia's Privacy Act reforms and the proposed AI transparency measures under the Albanese government's AI governance agenda don't yet specifically regulate real-time biometric surveillance in public spaces, leaving a significant regulatory gap.

The Mandarin ↗

AutoJack: Microsoft Researchers Show How a Web Page Can Hijack an AI Agent for Full Code Execution

Microsoft's own security team has detailed a new exploit chain called AutoJack that weaponises AI browsing agents against their host machines. The attack works by steering an AI agent to load a malicious web page; JavaScript on that page then reaches a privileged local service running on the same machine and spawns a process with host-level access. No credentials required, no additional user interaction after the initial navigation. It's a clean demonstration of why AI agents — which browse, execute, and interact with systems on your behalf — represent a fundamentally different threat surface than a traditional application. The finding underscores the identity and access governance problem that comes with agentic AI: the agent is trusted, so whatever it touches inherits that trust.

The Hacker News ↗

Splunk Enterprise Flaw Under Active Exploit — CISA Gives Agencies Until Sunday to Patch

CISA has added a critical Splunk Enterprise vulnerability, CVE-2026-20253, to its Known Exploited Vulnerabilities catalogue and given US federal agencies until Sunday to patch — one of the shortest remediation windows the agency has ever issued. The flaw allows unauthenticated remote code execution, making it especially dangerous given Splunk's role as the logging and SIEM backbone for many large organisations. Attackers began exploiting it within days of public disclosure, which is an increasingly common pattern as threat actors invest in rapid weaponisation pipelines. Splunk is widely deployed across Australian enterprise and government environments; organisations running it should treat this as a weekend priority regardless of CISA's jurisdiction.

SecurityWeek ↗

Operation Endgame Dismantles SocGholish — 106 Servers Down, 15,000 WordPress Sites Cleaned

An international law enforcement operation involving Dutch, German, Canadian, and US authorities has disrupted the SocGholish botnet, taking down 106 command-and-control servers and remediating nearly 15,000 infected WordPress sites. SocGholish, linked to the Russia-based cybercrime group Evil Corp, typically works by injecting fake browser update prompts into legitimate websites — visitors are tricked into running malware. The takedown is a meaningful operational disruption, though Evil Corp has historically rebuilt after law enforcement actions. The sheer number of infected WordPress sites cleaned up is notable: SocGholish's distribution model relied on compromising trusted sites, making it harder for users and security tools to detect the initial infection vector.

The Hacker News ↗

Klue Supply Chain Attack Hits Huntress, Recorded Future via Salesforce OAuth Abuse

A supply chain attack against competitive intelligence platform Klue has had downstream consequences for several cybersecurity companies, including Huntress and Recorded Future. Attackers abused OAuth tokens to access Salesforce instances belonging to Klue's customers, exfiltrating data before Salesforce disabled the integration. The incident is a useful case study in third-party OAuth risk: the breach didn't require compromising Salesforce directly — it went through a connected application that had been granted broad permissions. Salesforce has blocked the Klue Battlecards integration while the investigation continues. For Australian organisations, third-party app integrations with Salesforce are common, and this incident reinforces the ACSC's guidance on reviewing and limiting OAuth application permissions in SaaS environments.

SecurityWeek ↗

Gentlemen RaaS Deploys GentleKiller Framework Targeting 400 Security Processes

The Gentlemen ransomware-as-a-service operation has built and maintains a dedicated EDR-killing framework called GentleKiller, which affiliates receive as part of their attack toolkit. The framework targets more than 400 distinct security processes — endpoint detection tools, AV engines, logging agents — and supplements its own code with third-party EDR-killer utilities. The operational sophistication here is the notable detail: this isn't a one-off tool but an actively developed, version-controlled framework handed to affiliates like enterprise software. It means defenders can't rely on detecting a single known EDR-killer binary; they need to watch for the behaviour of security processes being terminated rather than the specific tool doing the terminating.

The Hacker News ↗

Startup Claims It Solved the Mathematical Bottleneck Holding Back LLMs

Miami-based AI startup Subquadratic emerged from stealth last month claiming it had cracked a core mathematical limitation in transformer architectures — the quadratic scaling problem that makes attention mechanisms increasingly expensive as context windows grow. Initial scepticism was high, given the thinness of the technical details at launch. This week the company began sharing more substantive evidence. If the claims hold up to peer scrutiny, it could meaningfully reduce the compute cost of running large context-window models, with implications for inference costs across the industry. The AI infrastructure investment wave — exemplified by Baseten reportedly raising $1.5 billion — suggests the market is very ready for cheaper inference.

MIT Technology Review ↗

Cisco Acquires WideField Security to Extend Splunk's Agentic SOC Ambitions

Cisco has announced the acquisition of WideField Security, a startup focused on identity, credential, and session analysis for security operations. The deal is explicitly framed around bolstering Splunk's agentic SOC roadmap — the idea that AI agents, rather than analysts, will handle much of the initial triage and investigation work in a modern security operations centre. WideField's technology is meant to give those agents broader visibility into blast radius when an account is compromised: not just what happened, but what the attacker could reach from there. It's a strategic signal about where Cisco sees the SOC evolving, and it follows Accenture's $4.1 billion OT security bet earlier this week in suggesting the enterprise security market is consolidating fast.

SecurityWeek ↗

Schneier: The Anthropic Ban Misses the Point Entirely

Security researcher Bruce Schneier has published a sharp critique of the US government's decision to classify Anthropic's Fable model as a dangerous munition and force it offline. His core argument: the threat isn't any single model, it's the broader trajectory of AI capability development — and export controls applied to one company's models do nothing to slow that trajectory. Other labs, open-source projects, and foreign governments are building equivalent capability anyway. Schneier's piece is worth reading alongside the Anthropic story as a dissenting view on whether supply-side model restrictions are a coherent security policy at all, or just security theatre that harms legitimate users without meaningfully reducing risk.

Schneier on Security ↗

The ASML Chip Tool Dispute: US Says It's in China, ASML Says It Isn't

A dispute has emerged between the US government and Dutch chip equipment manufacturer ASML over whether one of ASML's most advanced extreme ultraviolet lithography machines has ended up in China, in potential violation of export controls. ASML is pushing back, arguing the commercial logic alone makes it implausible the company would risk its export licence — the consequences of a violation would be existential for its US business. The standoff reflects the broader tension in semiconductor export controls: the US is trying to enforce restrictions on technology it doesn't manufacture, through companies in allied nations that have their own commercial interests. Australia's own export control framework under the Defence Export Controls office watches these precedents closely.

TechCrunch ↗

India Bans Telegram, Users Rush to VPNs

India has banned Telegram, prompting a surge in VPN downloads and migration toward rival messaging platforms. Telegram has responded by arguing that platform-level bans are a disproportionate response — the company's position is that governments should work with it to remove specific harmful content rather than blocking access entirely for all users. India's action follows a pattern seen in several countries of using platform bans as a blunt instrument for content moderation, and it raises familiar questions about the effectiveness of such measures given how readily users circumvent them. Australia's Online Safety Act gives the eSafety Commissioner similar but narrower powers to issue blocking directions, making the India precedent worth watching.

TechCrunch ↗

Sources consulted