Anthropic's 'Claude Mythos' Leak Sparks AI Alarm

Tear down the walls, Tech Insiders.

Siri is finally letting Google's Gemini and Anthropic's Claude into the iPhone, and Hark wants to break AI out of your pocket entirely. But as Anthropic's server leak and the FBI's email hack remind us, when the doors swing open, anyone can stroll through.

Let's cross the threshold.

Here's what you need to know today:

Anthropic's Ironic Leak Unveils 'Claude Mythos' Cyberthreat

Fortune first uncovered a publicly exposed trove of 3,000 Anthropic draft assets, including a blog post touting a supersized model dubbed Claude Mythos (aka Capybara). Anthropic confirmed the model is a "step change" above Opus 4.6, smashing benchmarks in coding, academic reasoning, and—crucially—cybersecurity.

The company says Mythos is already in early-access testing but admits its power could outpace today's cyberdefense tools.

Investors heard "unprecedented cybersecurity risks" and ran for cover. The iShares Cybersecurity ETF slid by more than 3%, while heavyweights CrowdStrike and Palo Alto Networks sank 6% each, and Tenable plummeted nearly 10%.

Anthropic insists the leak was pure "human error" (an unsecured, passwordless CMS), which is rich irony for a company currently warning the world about next-gen cyberthreats.

The company has since locked the door on the errant CMS. Still, the fiasco hands rivals and regulators fresh ammo just as Anthropic eyes a 2026 IPO and wrangles with Washington over a recently blocked Trump administration ban on federal use of its tech.

Why it matters: If Mythos really can automate zero-day hunts, defenders face an AI arms race they didn't budget for. It's yet another reminder that today's security stacks might be yesterday's news.

Share this newsletter!

Pulse Check

Should Anthropic release Claude Mythos broadly this year?

Results from Friday's Pulse Check

Would you let Gemini answer your next customer-service call?

Insider Intel

Siri Opens Doors to Gemini, Claude in iOS 27

Bloomberg says Apple's upcoming iOS 27 will split Siri's monogamous ChatGPT marriage, letting third-party AI chatbots like Google Gemini, Anthropic Claude, and any other App Store suitor answer your voice requests. A new "Extensions" menu will let you choose which brain handles coding help, trivia, or homework excuses.

The plug-and-play design means Apple skips future one-off integration deals and skims its usual up to 30% cut from any paid AI subscriptions activated through the store. (Apple already struck a separate, underlying deal for Gemini to power Siri's core after Anthropic reportedly demanded several billion dollars a year for the gig.)

Analysts see the move as both a catch-up play (Siri still trails Alexa and Google Gemini in flexibility) and a new source of services revenue. Internally, execs reportedly sparred over OpenAI's "staying power," and former AI chief John Giannandrea pushed for Google from day one.

Expect the Extensions kit to headline WWDC on June 8, alongside a standalone Siri app and new "Ask Siri" interface touches.

Flash Bytes

Why AI Infrastructure Has to Change

AI infrastructure isn't underpowered; it's underutilized. Hosted·ai fixes that with a deploy-anywhere GPU cloud stack for AI—combining compute, storage, and GPU optimization to make AI efficient and affordable. No idle GPUs. No brakes on innovation. No inflated bills. Just fully utilized, ultra-efficient infrastructure built for AI.

Security Alerts

Iran Hackers Leak FBI Chief's Old Gmail

Iran-aligned Handala Hack Team successfully breached FBI Director Kash Patel's personal Gmail, posting 300-plus emails and candid photos from 2010–2022. The FBI confirms the hit, noting that no classified data was involved—just "historical" personal files—and that it's chasing the culprits.

Analysts call this psychological warfare: Handala recently wiped tens of thousands of Stryker devices and doxed Lockheed staff to rattle US resolve as the Iran war grinds on.

The State Department's Rewards for Justice program is offering a bounty of up to $10 million.

If the FBI boss can get his personal email hacked, your inbox is definitely at risk. Protect yourself by turning on hardware-key MFA, disabling auto-forwarding, and tidying up old inbox archives—before your vacation selfies headline Reuters.

iOS 26.4 Forces UK Age Proof Checks

Apple's iOS 26.4 now blocks UK iPhone users from adult-rated apps, unblurred nudes in Messages, and unrestricted web browsing until they prove they're 18+, a first for Europe.

The change, nudged by the UK's Online Safety Act, appears as a post-update prompt and quietly enables content filters for anyone who can't (or won't) comply.

Adults breeze through if Apple already sees a credit card or an ancient Apple account history. Otherwise, you must scan a driver's licence, scan a PASS-accredited card, or add a credit card (debit cards won't work). Skip the step, and you'll still call Mum, but dating and adult sites stay behind the velvet rope.

Industry Shakeups

Figure AI Founder Launches Hark for Home AI Gadgets

Billionaire Brett Adcock is funding a new venture called Hark to build a "family" of personal-AI devices that live with you and around your home. You might know him as the guy behind Archer's flying taxis and Figure's humanoid robots (which last week made a bizarre White House cameo with Melania Trump).

After eight months in stealth, Hark is officially opening applications for its beta platform.

Backed by $100 million of his own cash and a dream team poached from Apple, Meta, and Google, Hark is developing its own multimodal language model tuned for speech, vision, and long-term memory, slated to debut this summer, with hardware following "soon after."

Adcock says phones and wearables are "pre-AI" relics. Hark's yet-to-be-revealed gadgets will proactively anticipate calendars, habits, and even emotional cues. Just don't expect a smart pin: Hark's lead designer, ex-Apple alum Abidur Chowdhury, has voiced heavy skepticism about today's camera pins and clip-on wearables.

Hark is wading into a shark tank: OpenAI and Jony Ive are crafting their own mystery device, Meta is doubling down on smart glasses, and Apple is prototyping an AI pin and camera-equipped AirPods. But Adcock insists there's room for multiple form factors, hinting that Hark's portfolio will stretch from pocket to living room.

With 50 staff today and a 150-person headcount target by summer, the startup is installing thousands of Nvidia B200 GPUs to train its models—and even borrowing Figure's robots as testbeds for "agentic" AI behavior (hopefully without the skull-fracturing force alleged in a November 2025 whistleblower lawsuit).

Whether Hark can escape the fate of Humane's Pin or Rabbit's R1 hinges on delivering real utility, not just hype, before the smartphone swallows the category.

Meet Our Author

Justin Meyers

Writer at TechnologyAdvice

Justin Meyers is an investigative writer and editor who draws on over a decade of meticulous hands-on research to deliver the full, trustworthy story behind consumer and enterprise tech, including cybersecurity.

Curious about where AI is really headed?

The Neuron cuts through the noise to bring you smart, hype-free takes on the latest AI trends, tools, and breakthroughs. Join 500,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

Advertise in Daily Tech Insider!

Daily Tech Insider is a TechnologyAdvice business.

© 2026 TechnologyAdvice, LLC. All rights reserved.

TechnologyAdvice, 3343 Perimeter Hill Dr., Suite 215, Nashville, TN 37211, USA.