Anthropic flags self-improving AI risk; Marvell agrees $115B Lumentum deal

Anthropic's IPO filing warns of AI-accelerated development

Anthropic filed an amended S-1 on Monday 2 November (US time). In its risk factors, it says internal estimates "indicate that we may imminently meet" the point in its Responsible Scaling Policy defined as "dramatically accelerated progress attributable to AI". No current model has been designated as crossing that line, and the company says it's still assessing the data. It also admits it has missed big capability jumps in the past and is considering policy changes to keep humans in control of frontier model development.

The filing shows third-quarter revenue of about $17.0 billion, up from $1.38 billion a year earlier. The quarter ended with an operating loss of $567 million, after operating income of $974 million in the second quarter. Anthropic puts the loss down mainly to $5.0 billion of compute bought on short-term commitments, plus $2.0 billion in non-cash charges from its staff donation-matching program. An S-1 is partly a sales document, so the safety language is likely to be debated.

What it means for NZ organisations: Claude is now backed by disclosed, audited-style numbers, and that helps with vendor risk reviews. Boards relying on frontier models should keep an eye on how Anthropic changes its scaling policy, because those changes could affect model release timing and access.

Marvell agrees to buy Lumentum for $115 billion

Marvell Technology announced on 2 November an all-stock deal for Lumentum, the Nvidia-backed maker of lasers and optical parts for AI data centres. Lumentum shareholders will get 4.5 Marvell shares for each share they hold, about $1,132 per share and a premium of roughly 23.9%. The deal values Lumentum at $115 billion including debt. Marvell holders will own about 67% of the combined company. The companies expect over $1 billion a year in cost savings and plan to close in the first half of 2028, subject to shareholder and regulatory approval.

The reasoning behind the deal is that AI clusters are hitting the physical limits of copper wiring and moving to optical links.

What it means for NZ organisations: Optical networking is becoming a bottleneck in AI data centres. Local data centre builders and cloud buyers should expect supplier consolidation and plan around longer lead times for this equipment.

US publishers sue Brave over AI scraping

The New York Times sued Brave Software in Manhattan federal court on 2 November. Hours later, the publishers of the New York Daily News, Chicago Tribune and Orlando Sentinel filed a separate suit. They allege Brave scraped their journalism into an index of more than 30 billion pages and sold access through its Search API to AI companies, naming Anthropic, Perplexity, Mistral, xAI and Cohere as customers. They also say Brave's crawler has never been identified, so sites can't block it. None of these claims has been tested in court.

What it means for NZ organisations: If you buy AI search or retrieval services, ask the provider where its data comes from. Publishers here should check whether their robots.txt rules actually cover every crawler that's reading their sites.

UK cyber agency expects AI attacks on critical infrastructure by 2028

On 3 November, Richard Horne, head of the UK's National Cyber Security Centre, said at the launch of its annual report that the UK will likely face AI-enabled cyberattacks on critical infrastructure by 2028. He said AI is more likely to scale up existing tactics, such as ransomware that finds and exploits known weaknesses, than to discover entirely new ones. Malicious cyber activity costs UK organisations almost $20 billion a year.

What it means for NZ organisations: Operators of power, water and manufacturing systems should assume old, unpatched vulnerabilities will be found faster. The NCSC's basics still apply: patch promptly, limit admin access and keep tested backups.

OpenAI pitches "compute campuses" and a federal testing bill

Semafor reported on 3 November that OpenAI is reworking its policy message after the US midterms. The company is renaming data centres "compute campuses", asking communities for input on new builds and pushing for global AI standards. Chris Lehane, OpenAI's chief lobbyist, said the company will push for a mandatory AI testing bill that he thinks could win 60 Senate votes. OpenAI still backs two California frontier AI safety bills. Brad Carson of Americans for Responsible Innovation called the renaming "Orwellian".

What it means for NZ organisations: Mandatory pre-release testing in the US could become a template that other countries copy. Firms writing AI governance policies should expect testing evidence to become a standard thing to ask vendors for.

OpenAI hires former US Army Secretary

CNBC reported on 2 November that OpenAI has hired Dan Driscoll, who left the role of US Army Secretary earlier this year, as senior counselor for national security modernization. It's a new role. He'll advise leadership on AI for national security and cybersecurity, including helping US allies adopt advanced AI, and his work builds on OpenAI's Daybreak cyber-resilience program.

What it means for NZ organisations: OpenAI is clearly courting allied governments. NZ agencies and defence suppliers can expect more offers aimed at government, and should weigh them against data sovereignty requirements.