Tuesday 21 April 2026

Tim Cook Is Out. What Happens to Apple Now?

Tim Cook announces his exit from Apple, Anthropic's Mythos model lands at the NSA, and a supply-chain hack at Context.AI cascades into a breach at Vercel — a busy 24 hours across tech and security.

Lead story

Tim Cook Is Out. What Happens to Apple Now?

After fourteen years running the most valuable company on earth, Tim Cook is stepping down as Apple's CEO. Hardware engineering chief John Ternus will take the reins on 1 September 2026, with Cook moving to executive chairman. The transition was announced Monday evening and is, by Apple standards, a remarkably clean handover — no drama, no sudden departure, just a tightly controlled press release and a summer of overlap.

Cook's tenure is almost impossible to overstate. He inherited a company riding the iPhone wave in 2011 and turned it into a $3+ trillion institution held together as much by supply chain discipline and services revenue as by any single product. The App Store, Apple Silicon, AirPods, the pivot to services — most of the things that define Apple today were built on his watch, not Steve Jobs'.

Ternus is a less obvious choice than the rumoured names — Eddy Cue, Craig Federighi — but arguably the right one for where Apple sits in 2026. He's spent his career in hardware, overseeing the M-series chip transition and the recent spatial computing push. If Apple's next act is genuinely physical — mixed reality, robotics, health hardware — a hardware engineer in the top job makes a certain kind of sense.

The harder question is what Cook leaves unfinished. Apple's AI story remains the shakiest part of its portfolio. Apple Intelligence has been more cautious rollout than transformative leap, and every competitor — Google, Microsoft, Samsung — is shipping faster and more aggressively. Ternus inherits a company with extraordinary financial health and a loyal user base, but one that is measurably behind in the technology most investors are currently pricing into valuations.

There's also the regulatory overhang. Apple faces ongoing antitrust scrutiny in the EU, the US, and multiple Asian markets. The App Store business model is under sustained legal attack. Cook was a skilled Washington and Brussels operator; Ternus has no public record in that arena. That gap could matter quickly.

What to watch: How Ternus handles the Apple Intelligence roadmap in his first few months will be the clearest signal of whether this is a genuine strategic pivot or a steady-as-she-goes succession. The September iPhone launch — his first as CEO — will be scrutinised in a way no Apple event has been since 2011. Cook stepping into an executive chairman role also means he's not gone; that dynamic, Jobs-era echoes aside, is worth watching carefully.

Also today

Vercel Hacked via Roblox-Themed Malware at Third-Party AI Tool

The breach at web infrastructure provider Vercel traces back to a surprisingly mundane origin: malware disguised as Roblox cheats infected a machine at Context.AI, a third-party analytics tool used by a Vercel employee. The attacker leveraged that foothold to hijack the employee's Google Workspace account, then pivoted into Vercel's internal systems. ShinyHunters subsequently claimed to be selling the stolen data for $2 million. The incident is a clean illustration of how SaaS integrations with over-privileged access turn a vendor's bad day into your breach. Vercel has confirmed limited customer credentials were exposed.

CyberScoop ↗

Anthropic's Mythos Is Already Inside the NSA

Anthropic's Mythos model — a restricted, high-capability AI system that has drawn concern from researchers over its potential to accelerate offensive cyber operations — is reportedly already in use by the NSA, despite an ongoing feud between Anthropic and the Pentagon over AI procurement. The deployment signals that intelligence agencies are moving fast to operationalise frontier AI regardless of which contractors hold the official contracts. Separately, Ars Technica reports that Mythos has sparked broader alarm among security researchers who worry that cyberdefences simply cannot be patched and deployed fast enough to keep pace with AI-assisted exploit development.

TechCrunch ↗

Anthropic MCP Has a 'By Design' RCE Problem

Researchers have identified a critical architectural weakness in Anthropic's Model Context Protocol — the increasingly popular standard that lets AI agents connect to external tools and data sources. The flaw isn't a bug in a specific implementation; it's baked into the protocol's design, allowing a malicious MCP server to execute arbitrary commands on any system running a vulnerable client. The concern isn't just individual compromise — it's the supply chain implications. As MCP adoption spreads across AI toolchains, one poisoned server could cascade through a large number of downstream deployments. Anthropic has not yet indicated whether a protocol-level fix is planned.

The Hacker News ↗

Scattered Spider Ringleader Pleads Guilty in the US

Tyler Buchanan, a British national believed to have led the Scattered Spider cybercrime collective, has pleaded guilty in a US federal court to wire fraud and aggravated identity theft. He becomes the second Scattered Spider-linked individual to enter a guilty plea in the US. Scattered Spider is responsible for a string of high-profile intrusions, including the MGM Resorts and Caesars Entertainment attacks in 2023, and used SIM-swapping and social engineering as its primary entry vectors. Buchanan's guilty plea is a meaningful step in a case that has tested international law enforcement cooperation — the group operated largely from the UK and the US.

Bleeping Computer ↗

North Korean Hackers Blamed for $290M Kelp DAO Crypto Heist

A North Korean threat actor has been attributed with stealing nearly $290 million in cryptocurrency from Kelp DAO, making it the largest crypto theft of 2026 so far. The attack involved multiple prominent DeFi platforms and is consistent with the pattern of DPRK-linked groups using crypto theft to fund the state's sanctions-busting operations. North Korean hackers have now stolen billions from crypto platforms over the past several years, and their techniques — supply chain compromise, social engineering of developers, and bridge exploits — continue to evolve. The incident is currently under investigation by blockchain security firms.

The Record ↗

Microsoft Teams Is Being Weaponised for Helpdesk Fraud

Microsoft has issued a warning that threat actors are increasingly abusing Teams' external messaging capabilities to impersonate IT helpdesk staff. The attack pattern involves creating external Teams accounts that mimic internal support teams, then convincing employees to grant remote access or hand over credentials. Once inside, attackers use legitimate tools — RMM software, built-in Windows utilities — to move laterally, which makes detection harder. It's a social engineering attack dressed up in enterprise collaboration tooling, and it's working. Microsoft recommends restricting external Teams access and enabling alerts for unusual external contact.

Bleeping Computer ↗

ZionSiphon Malware Targets Israeli Water Infrastructure

Researchers at Darktrace have identified a previously unknown malware strain, dubbed ZionSiphon, that appears purpose-built to target Israeli water treatment and desalination systems. The malware establishes persistence, tampers with local configuration files, and scans for operational technology services on the local subnet — behaviour consistent with pre-positioning for a disruptive or destructive attack rather than simple espionage. Water infrastructure has been a recurring target in the broader Israel-Iran cyber conflict. Attribution hasn't been formally confirmed, but the targeting specificity and OT focus strongly suggests a state-sponsored actor.

The Hacker News ↗

Critical RCE in SGLang AI Serving Framework Scores CVSS 9.8

A severe vulnerability tracked as CVE-2026-5760 has been disclosed in SGLang, the open-source high-performance framework widely used to serve large language models in production environments. The flaw is a command injection issue that can be triggered by loading a malicious GGUF model file, resulting in arbitrary code execution on the host. With a CVSS score of 9.8, it sits at the top end of the severity scale. Given how broadly SGLang is deployed in AI inference infrastructure — often in cloud environments with significant downstream access — the blast radius of an unpatched instance could be substantial. Patch details are available in the disclosure.

The Hacker News ↗

Deezer: 44% of Daily Music Uploads Are AI-Generated

Streaming platform Deezer has revealed that roughly 75,000 AI-generated tracks are submitted to its platform every single day, making up 44 percent of all daily uploads. Despite the flood, actual AI music listening remains tiny — somewhere between one and three percent of total streams — and Deezer says it's demonetising 85 percent of those streams after detecting them as fraudulent. The platform is positioning its detection tooling as an industry benchmark. The numbers are striking less for what people are listening to and more for what they reveal: content farms are clearly automating music creation at scale to game streaming royalty systems.

Ars Technica ↗

TSMC Earnings Signal Cracks in the AI Chip Narrative

TSMC's latest quarterly earnings, dissected by Stratechery's Ben Thompson, reveal something subtle but important: TSMC's leadership doesn't appear fully bought into the AI hypergrowth story that has inflated much of the semiconductor sector's valuations. While revenue remains strong and N3 fab expansion is proceeding, the company's forward guidance and internal framing suggest caution about the durability of current AI infrastructure spending. For investors and builders betting on sustained AI compute demand, TSMC's temperature-taking matters — the company has better visibility into actual chip orders than almost anyone else in the stack.

Stratechery ↗

Network 'Background Noise' Can Predict Edge-Device Vulnerabilities Before They're Exploited

Researchers at GreyNoise have published findings suggesting that surges in internet background noise — the constant low-level scanning and probing traffic that washes over every internet-connected device — reliably precede the disclosure and active exploitation of vulnerabilities in edge devices like firewalls and VPN appliances. By monitoring for unusual upticks in traffic patterns targeting specific ports or device types, defenders may get an early-warning signal days before a CVE drops or an exploit campaign begins. If the methodology holds up, it could meaningfully shift the defender timeline — from reactive patching to proactive hardening.

CyberScoop ↗

Sources consulted