Tech news, 1 November 2026: a backdoored Claude Code extension, a month of AI platform breaches, Binance attribution and Qwen in Apple Intelligence

1. A tampered Claude Code extension tried to backdoor developer machines

Cursor says tampered versions 2.1.312 and 2.1.313 of the Claude Code extension were listed in its extension marketplace for a few hours on 29 October, and that up to about 5,000 users may have received them²³. Microsoft's analysis describes a loader that tried to add an attacker SSH key, enable the SSH server and open a reverse tunnel; it notes many attempts would have failed without administrator rights, and that the official Visual Studio Marketplace listing and npm package were not affected¹. Anthropic says its own systems were not compromised.

CTO take: AI coding tools run with wide access on developer machines, and auto-update makes a marketplace compromise reach people in hours. If your teams use Cursor, follow Cursor's remediation now: check for the planted key and persistence, and rebuild if in doubt. Longer term, pin and allow-list editor extensions, and keep production credentials off developer laptops, as covered in AI coding policy for engineering teams.

2. Wiz describes a "Black October" for AI developer platforms

A Wiz Research report published on 30 October describes a run of breaches at AI developer and hosting platforms during October, with stolen AI service credentials feeding LLMjacking, the resale of model access, and enterprise keys used for large volumes of model calls⁴. Coverage notes Wiz does not attribute all the incidents to one actor or one vulnerability⁴.

CTO take: AI API keys are now money in an attacker's hands. Rotate keys stored on third-party platforms, set spending limits and alerts per key, and route model access through your own gateway so one leaked credential does not mean unlimited spend. See AI cost allocation and keeping secrets out of state.

3. Analysts tie the Binance theft to North Korea

TRM Labs attributes the Binance theft to North Korea's TraderTraitor unit with high confidence, tracking about $6.36 billion leaving hot wallets across ten blockchains, and assesses that hot wallet signing infrastructure was compromised⁵. SecurityWeek reports Chainalysis reached a similar view; Binance and governments had not formally named an attacker at the time of writing⁶.

CTO take: state-backed groups target engineers' machines to reach signing and payment systems. The controls in the Binance hack: architecture lessons apply to any platform that moves money, from exchanges to payment companies.

4. Alibaba's Qwen will power Apple Intelligence in China

Apple and Alibaba said on 31 October that Qwen will power the China edition of Apple Intelligence, with a developer beta from 4 November and a public release planned for December; the rebuilt Siri shown earlier this year is not included in this release⁷.

CTO take: even the largest platforms now ship different AI models by market to meet local rules. If you build AI products for China or other regulated markets, design for a model per region behind one interface, with evaluations run per market. See model routing and where I work.

What I'd do this week

• If anyone uses Cursor, check for the backdoor indicators and follow Cursor's remediation steps.
• Allow-list editor extensions and turn off auto-update for tools with broad machine access.
• Rotate AI API keys held on third-party platforms and set spending alerts per key.
• Make sure no developer laptop holds credentials that can move money or reach production data directly.

Need a second pair of eyes on developer and AI supply-chain security? See fractional CTO work or analyse your challenge and bring the result to a call.