Daily · AI Agent Security and Compute Wars · October 3, 2026

Key points

Compute Supply Chain: Cerebras Loss and Chip Smuggling

SemiAnalysis reported on Wednesday that OpenAI will power its 'Ultrafast' mode for GPT-6.1 Sol with Nvidia GPUs instead of Cerebras hardware. This development triggered a significant sell-off in Cerebras, whose shares fell nearly 20% for the week to their lowest price since the May IPO, closing Friday at $166.43. The company's market capitalization dropped from $95 billion on its first trading day to just over $39 billion. Sam Altman stated that Cerebras remains a close partner, yet the shift underscores Nvidia's continued dominance in AI compute infrastructure.

Simultaneously, the US Department of Justice arrested Greg Lui, owner of Earthmade Computer Inc., on charges of conspiracy to violate export controls, outbound smuggling, and money laundering. The indictment alleges that between October 2023 and August 2026, Lui used false documentation to route servers containing Nvidia A100 and H100 GPUs through Malaysia and Singapore to a Chinese firm in Hangzhou. Prosecutors stated the scheme involved over $300 million in restricted hardware, with Earthmade receiving more than $176 million from Malaysian transshipment companies between January and October 2024. Nvidia confirmed it is cooperating with law enforcement, emphasizing that smuggling is a losing proposition legally and economically.

These events highlight the tightening of both commercial and regulatory controls over AI compute. For infrastructure procurement, the reliance on Nvidia for high-speed inference remains paramount, while compliance teams must note the DOJ's active enforcement posture against export control violations in the GPU supply chain.

AI Agent Privacy: Muse Backlash and Apple Response

Meta's personal AI agent, Muse, launched in the US and Canada with a free tier and paid plans at $20 and $100 per month. However, the rollout has been marred by security and privacy concerns. A researcher discovered a local privilege escalation vulnerability in the Mac app two days before its public presentation, which Meta patched with a hotfix. More significantly, independent researcher Karan Joshi extracted Muse's internal files, revealing that the agent creates detailed profiles of every person in a user's life on an hourly basis, including relationship history and suggestions for strengthening connections.

In response to growing scrutiny over how AI agents access personal data, Apple announced it will introduce additional controls for Full Disk Access in macOS. The company stated that some developers are using this permission to expose files, mail, and messages without full user knowledge, a risk that grows as AI agents become more autonomous. Meta CTO David Singleton maintained that Muse's Messages integration is opt-in and requires explicit system-level permissions, but security expert Patrick Wardle argued that Full Disk Access allows reading any non-root file, including chats and browsing history.

The tension between agent capabilities and user privacy is now a central product challenge. As platforms like Airbnb signal reluctance to allow outside agents to make bookings, the definition of secure, consent-based agent interaction becomes a critical differentiator for consumer-facing AI products.

Data Center Opposition: Europe and Asia

Public opposition to data centers is spreading beyond the US, impacting investment timelines in Europe and Asia. Research from STL Partners indicates that public backlash has affected around $42 billion of data center investments in Europe through delays and cancellations, compared to $77 billion in the US. More than 70 projects in Europe were rejected or restricted between January and April 2026, a higher number than in all of 2025.

In Scotland, planning approvals for new hyperscale data centers have been paused after campaigners cited Ireland's power demand issues as a cautionary tale. Denmark passed an emergency law placing data centers at the back of the queue for grid power applications, while Spain proposed rules requiring 80% renewable electricity sourcing. In South Korea, residents in Geumcheon have protested for 172 days against a local data center, leading officials to plan a requirement for majority resident consent within 200 meters of proposed sites.

Amazon responded to the US backlash by committing to donate over $1 billion to communities neighboring its data centers over five years and pledging to be water positive by 2030. However, environmental group Stand.Earth criticized the move as damage control, noting that Amazon's commitments do not address the climate impact of a power plant expected to become a major source of US pollution.

US AI Governance: The 'AI Tsar' and Safety Debates

US President Donald Trump is expected to name Director of National Intelligence Jay Clayton as his top AI adviser, a role previously held by David Sacks. Clayton already oversees 18 US intelligence agencies and is expected to retain that position. This consolidation links AI policy directly to the intelligence apparatus, potentially shaping export controls and competitive posture toward China. Trump also signed an executive order this week renaming artificial intelligence 'super intelligence.'

The appointment comes amid public debate over AI safety. Yann LeCun, who founded AMI Labs after leaving Meta, stated he has 'zero concerns' about AI extinction and called Anthropic CEO Dario Amodei 'deluded,' arguing that recent rogue AI incidents are due to poor human oversight rather than inherent danger. Conversely, David Robinson, a former OpenAI safety lead, wrote in The Atlantic that OpenAI is failing to achieve the necessary level of care, suggesting AI firms should operate with layers of redundancy similar to nuclear power plants. Anthropic's Amodei called for slowing down the development of cutting-edge models and bringing on third-party evaluators, a plan publicly endorsed by Sam Altman.

Market Moves: Nvidia, Tesla, and Tokenization

Nvidia shares closed up 1.3% Friday, reaching a market capitalization of roughly $5.6 trillion, less than $400 billion from becoming the first company to hit $6 trillion. The company announced a record $150 billion increase to its buyback plan and unveiled a new AI security system designed to stop agents from going awry. Tesla plans to expand its Cybercab driverless ride-hail service beyond Austin to San Antonio, Dallas, Nevada, and Florida after a rocky first month, during which the fleet grew from 45 to 169 authorized vehicles.

In finance, BlackRock developed three tokenized investment portfolios for Ondo Finance, packaging high-income and growth strategies into single blockchain tokens. This move signals a shift toward onchain portfolio management, with model portfolios holding about $9.8 trillion in assets as of June. Meanwhile, EQT acquired a majority share in Swiss cybersecurity company Acronis at a valuation exceeding $3.5 billion, indicating continued consolidation in the European security market.

Our read

The shift of OpenAI's inference workloads to Nvidia GPUs confirms that despite new entrants like Cerebras, the compute supply chain remains heavily concentrated, raising the stakes for export control compliance as seen in the $300 million smuggling case. The privacy backlash against Meta's Muse and Apple's subsequent tightening of macOS permissions suggest that 'opt-in' agent access is no longer sufficient; explicit, granular user consent mechanisms will become a regulatory requirement rather than a best practice. The spread of data center opposition to Europe and Asia implies that infrastructure build-out timelines must now account for local community engagement and renewable energy mandates, potentially increasing capital expenditure and permitting risks.

This material was produced automatically by a large-language-model system from the public sources listed below; it is AI-generated content and may contain inaccuracies — verify facts against the original sources.

Sources