Feb 9, 2026

AI News Roundup: AI Takes Over Super Bowl LX as Anthropic and OpenAI Trade Blows

Super Bowl LX wasn’t just the biggest game of the year — it was the biggest AI advertising event in history. Nearly a quarter of all commercials featured AI in some capacity, with Anthropic and OpenAI going at each other directly on the biggest stage in advertising. Meanwhile, a $10 billion infrastructure deal landed in Australia, the EU escalated its fight with Meta over AI exclusivity on WhatsApp, and MIT researchers revealed that the LLM ranking platforms we all rely on can be flipped by removing just two votes.

Here’s everything that matters from February 9, 2026.

The Big Story: Anthropic and OpenAI Clash at Super Bowl LX

Anthropic debuted its first-ever Super Bowl campaign during Super Bowl LX, and it came out swinging. The company ran a 60-second pregame spot and a 30-second in-game ad with titles like “Deception,” “Betrayal,” “Treachery,” and “Violation” — all carrying the tagline “Ads are coming to AI. But not to Claude.” The message was unmistakable: a direct shot at OpenAI’s plans to sell advertising inside ChatGPT.

OpenAI didn’t stay quiet. The company countered with its own ad showcasing its Codex tool, and CEO Sam Altman pushed back on social media, calling Anthropic’s ads “funny” but “dishonest,” defending ads in ChatGPT’s Free and Go tiers as necessary for broad accessibility.

The clash is more than Super Bowl theater. It reflects a genuine strategic divergence between the two leading AI companies. Anthropic is betting that users will pay a premium for an ad-free experience and that trust is a competitive moat. OpenAI is betting that subsidized access via advertising is the path to mass adoption. Both strategies have precedent — streaming went through the exact same split — but the stakes are higher here because AI assistants handle sensitive personal and business data in ways Netflix never did.

The broader picture: 15 out of 66 Super Bowl LX commercials — nearly 23% — featured AI, with Google, Amazon, Salesforce, Meta, and others all running spots. Svedka even ran what’s being called the first “primarily” AI-generated national Super Bowl commercial. Whether you loved it or were sick of it, AI’s takeover of the ad break was impossible to ignore.

Today’s Top Stories

Firmus Technologies Secures $10B Blackstone-Led Loan

Nvidia-backed Australian AI infrastructure startup Firmus Technologies secured a $10 billion debt financing package led by Blackstone with participation from Coatue Capital — one of the largest private credit deals in Australian history. The funds will build the next phase of Project Southgate, developing AI training and inference infrastructure across Australia with capacity up to 1.6 gigawatts by 2028. Firmus is expected to IPO in Sydney later this year. The deal is another data point showing that AI infrastructure spending isn’t just a US phenomenon — it’s going global at breakneck speed.

EU Threatens Interim Measures Against Meta Over WhatsApp AI

The European Commission announced it has told Meta that it breached EU antitrust rules by restricting third-party AI assistants on WhatsApp. Since January 15, Meta has allowed only its own Meta AI assistant to operate within WhatsApp, blocking competitors. EU Competition Chief Teresa Ribera said the Commission is “considering quickly imposing interim measures on Meta, to preserve access for competitors to WhatsApp while the investigation is ongoing.” Meta rejected the charges. This could set a major precedent for whether platform owners can bundle their own AI into messaging apps while locking out rivals.

AI.com Launches at Super Bowl with Record $85M Campaign

Crypto.com founder Kris Marszalek debuted AI.com during Super Bowl LX after purchasing the domain for a record-breaking $70 million (paid entirely in crypto). The total campaign cost hit approximately $85 million. It worked — AI.com was named the most effective advertiser of the game, generating 9.1x as much impact as the median Super Bowl LX ad. There was just one problem: the servers crashed under the traffic surge.

MIT Finds LLM Ranking Platforms Are Alarmingly Fragile

MIT researchers discovered that LLM ranking platforms based on crowdsourced voting can be flipped by removing just 2 votes out of 57,000 — that’s 0.0035% of total votes. Even a more robust platform using expert annotators required removing only 83 of 2,575 evaluations (~3%) to change the top-ranked model. The finding has implications for anyone making model selection decisions based on leaderboard rankings and raises questions about how much we should trust crowdsourced AI benchmarks.

Claude Opus 4.6 Surges in Prediction Markets

Following its February 5 launch, Anthropic’s Claude Opus 4.6 has sent the company’s odds of holding “Best AI Model” surging from 40% to 68% in prediction markets. The model reportedly identified more than 500 previously undisclosed zero-day security vulnerabilities in open-source libraries during testing — a capability that, if sustained, could make it a significant tool for cybersecurity.

Quick Hits

AI Safety: The second International AI Safety Report 2026 was published, providing a comprehensive review of the latest scientific research on capabilities and risks of general-purpose AI systems.
AI Regulation: The Future of Life Institute launched an $8 million “Protect What’s Human” ad campaign urging stronger AI regulation, backed by Megan Garcia who sued Google and Character.AI after her son’s death following interactions with an AI chatbot.
SaaSpocalypse: Anthropic’s Claude plugins have now wiped $285 billion from SaaS market value as AI features increasingly resemble built-in platform capabilities rather than paid add-ons.
Semiconductors: TSMC committed to manufacturing advanced AI semiconductors in Japan, expanding the AI chip supply chain beyond Taiwan and the US.
Nvidia: Shares jumped 7.7% after CEO Jensen Huang said the AI infrastructure buildout still has 7-8 years to run and that GPUs sold six years ago are now increasing in value.
Funding: SambaNova raised $350 million led by Vista Equity Partners to position as a Nvidia alternative; ElevenLabs tripled its valuation to $11 billion; and Gather AI raised $40M for AI-powered warehouse drones.
Research: New arXiv papers found that AI agents show systematic overconfidence (22% actual success vs. 77% estimated), and that LLM reasoning decays exponentially with execution length.
HBR: Harvard Business Review argued that AI doesn’t reduce work — it intensifies it, applying Jevons Paradox to labor and warning about cognitive overload from managing AI agents.

What This Means

Super Bowl LX crystallized something that’s been building for months: AI is no longer a technology story confined to developer blogs and earnings calls — it’s a mass consumer narrative. When nearly a quarter of the most expensive ad slots in the world are devoted to AI products, the industry has crossed a cultural threshold. The Anthropic-OpenAI clash specifically highlights a defining strategic question for the next era: will AI be funded by subscriptions or by advertising, and what are the privacy and trust implications of each path? Meanwhile, the $10 billion Firmus deal, TSMC’s Japan commitment, and Nvidia’s 7-8 year buildout timeline all reinforce that the infrastructure layer is being built for a much longer cycle than any single model generation. And MIT’s finding that LLM rankings can be swayed by two votes should give everyone pause about how we’re making decisions about which models to trust.