Burgundy, Oct 20
Autonomous hacking agents, a DeepSeek ban and stolen API keys led a day when AI's risks and its balance sheet were both on display.
Security ran the day. The AI security firm Norforge says one hacker broke into more than 100 websites with an army of AI agents that mostly operated on their own, according to The Verge. Reuters reported the victims included John Deere, SharkNinja and Cummins. Gizmodo covered the same campaign, called the attacker a lone hacker, and put the number of compromised sites in the hundreds.
Then another breach. Hackers hit OpenCode Console, the online platform of an open-source coding agent, took some users' API keys and ran up charges on their AI accounts. TechCrunch reported that 1.4% of users were affected, some charged hundreds of dollars and a few more than $1,000. Co-founder Dax Raad said the breach hit the Console, not the coding tool itself, whose keys never reach the company. David Sacks said frontier models from Anthropic, Google and OpenAI refused to analyze the exploits because the payloads tripped their safety filters.
Governments and labs had their say. The US Department of Homeland Security banned DeepSeek across federal civilian agencies, including third-party services built on its models. Gizmodo reported that DHS fears the Chinese lab could send sensitive data to Beijing, and that its sign-up process collects personally identifiable information and may track and fingerprint users. Anthropic's red team warned that self-replicating 'AI worms' could become feasible by year's end.
The money and chips moved too. Amazon released TorchNeuron, which lets PyTorch apps run on its Trainium chips with a single configuration change and no code rewrite. TechCrunch called it Amazon's biggest shot at Nvidia yet, noting that about 85% of AI apps are written with PyTorch and citing Deepgram as processing 70% more audio per dollar on Trainium. Nvidia launched three open Nemotron 3.5 models for running agents in data centers and on PCs. Chip startup Tenstorrent raised $1.5 billion at a $6.2 billion valuation and chose to stay independent.
Doubts about the build-out surfaced. Microsoft pulled out of Nscale's planned Essex datacentre, once called the UK's largest, over power supply concerns, leaving the project without its anchor tenant just as the company was expecting to go public. Gizmodo wrote that AI is looking less and less like a sure thing for investors. The Verge reported that CoreWeave's bonds slipped into distressed territory last week, and argued its debt is becoming the whole AI industry's problem.
Closer to daily life, the Tech Transparency Project found that test accounts posing as 13-year-olds could download 135 apps able to generate nonconsensual nude imagery, despite Apple and Google's parental controls. The Human Technology Institute warned that AI is speeding up workplace monitoring in Australia, often without staff being told. And music publishers said AI had been trained on Kylie Minogue and Nick Cave songs without payment, urging Australia to resist what they called strong-arming by tech firms over copyright.
The legal fallout reached OpenAI itself. A judge ordered former engineer Abdulrahman Alfozan, accused of spying on executives, to remain in custody, with the FBI saying he kept a folder called 'Saudi Gov Docs.'
Sources
- A hacker broke into John Deere and SharkNinja with an army of AI agents · The Verge, AI
- The Fear of a One-Man-Band Weaponizing AI Agents Is Becoming Reality · Gizmodo, AI
- Trump Admin Bans DeepSeek From Federal Agencies · Gizmodo, AI
- Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool · TechCrunch, AI
- Hackers hit AI coding tool OpenCode, running up charges on users’ AI accounts · TechCrunch, AI
- Nvidia launches new open models to run agents in the enterprise and on PCs · TechCrunch, AI
- Microsoft pulls out of AI datacentre project in Essex over power supply concerns · The Guardian, Artificial intelligence (AI)
- Microsoft Pulls Out of Giant British AI Cloud Deal Just As Cloud Company Was Expecting to IPO · Gizmodo, AI
- Tenstorrent hits $6.2B valuation to scale open-weight AI chips · TechCrunch, AI
- CoreWeave’s debt is becoming the whole AI industry’s problem · The Verge, AI
- Accused OpenAI spy had folder called “Saudi Gov Docs,” FBI says · Ars Technica, AI
- What are self-replicating ‘AI worms’, and why is Anthropic worried about them? · The Guardian, Artificial intelligence (AI)
- US teens can access ‘nudify’ apps despite Apple and Google’s parental controls, report finds · The Guardian, Artificial intelligence (AI)
- Calls to strengthen workplace privacy in Australia as employers turn to AI to track staff · The Guardian, Artificial intelligence (AI)
- Kylie Minogue songs already used to train AI as tech firms ‘strong-arm’ Australian government, music body warns · The Guardian, Artificial intelligence (AI)