Wednesday the 3rd of December

G’day folks,

It’s Tuesday the 3rd of December and the industry’s biggest players are locked in a fierce battle for dominance—with OpenAI declaring a Code Red over Google’s surging Gemini 3.

At the same time, Microsoft is quietly pumping the brakes on AI sales expectations as enterprise customers demand proof of ROI, not just fancy demos—raising serious questions about whether the AI boom can sustain its momentum.

Let’s get into it.

This newsletter is powered by:

• Letz - Create what you love

• Firecrawl - The Web Data API for AI

• Anything - Turn your words into mobile apps, sites, tools, and products

• Browserbase - A web browser for your AI

• 9x - Upskill your teams in AI

• Blackbox AI - Run multiple AI-agent simultaneously

In a Nutshell

• AI Safety Crisis: Major AI companies including OpenAI, Meta, Anthropic, and xAI are “far short of emerging global standards” in safety practices, according to a new Future of Life Institute report released today—with none having credible plans to prevent catastrophic AI risks.

• OpenAI’s Code Red: Sam Altman issued an internal “Code Red” memo to OpenAI staff yesterday, marshaling resources toward improving ChatGPT as Google’s Gemini 3 gains serious ground with 650 million monthly users and strong benchmark results.

• Microsoft’s AI Reality Check: Microsoft quietly slashed AI software sales quotas this week after salespeople missed growth targets, as enterprise customers resist premium pricing without proven ROI—raising serious questions about the $3.5 trillion AI bubble.

• Nvidia’s Open AI Breakthrough: Nvidia released Alpamayo-R1 at NeurIPS this week, the world’s first industry-scale open reasoning model for autonomous vehicles that lets self-driving cars “think out loud” about their decisions.

Top Stories

🚨 AI Companies Fail Safety Standards Across the Board

Eight leading AI companies—including OpenAI, Meta, Anthropic, xAI, and DeepSeek—are failing to meet emerging global safety standards, with none possessing robust strategies for controlling advanced AI systems, according to a bombshell report released today by the Future of Life Institute.

The details:

• The independent expert panel found that while companies race to develop superintelligence, they lack credible plans to prevent catastrophic risks.

• The findings come amid heightened public concern after several cases of suicide and self-harm were tied to AI chatbots.

• “Despite recent uproar over AI-powered hacking and AI driving people to psychosis and self-harm, US AI companies remain less regulated than restaurants and continue lobbying against binding safety standards,” said Max Tegmark, MIT professor and Future of Life president.

Why it matters: As AI systems become more powerful and autonomous, the gap between capabilities and safety measures is widening dangerously. The report validates concerns from AI pioneers like Geoffrey Hinton and Yoshua Bengio, who called for a ban on superintelligent AI development until science paves a safe path forward. With companies racing toward AGI, the absence of meaningful safety protocols is becoming impossible to ignore. Reuters

⚔️ OpenAI Declares Code Red as Google Surges Back

Three years after Google sounded its own Code Red over ChatGPT, the tables have turned. Sam Altman issued an internal memo yesterday declaring a Code Red to marshal resources toward improving ChatGPT, as competitive pressure from Google and other rivals intensifies.

The details:

• Google’s Gemini 3 launched two weeks ago with a confident, widespread rollout across Google’s ecosystem reaching billions of users—its fastest-ever deployment into Search.

• Gemini now boasts 650 million monthly users (October data), and some reports suggest OpenAI lost 6% of its users in the week following Gemini 3’s launch.

• As part of the shift, OpenAI is delaying other initiatives, including advertising plans. “We are at a critical time for ChatGPT,” Altman wrote, warning staff to expect “rough vibes” ahead.

• Altman said OpenAI will release a new reasoning model next week that beats Gemini 3 in internal evaluations.

Why it matters: The AI arms race has come full circle. Google, which invented the transformer architecture that powers every modern LLM and was once considered the global leader in AI research, lost its crown to OpenAI’s ChatGPT moment in late 2022. Now it’s clawing back with force. For OpenAI, which needs to raise an additional $100 billion while burning through cash, maintaining its edge is existential—especially after dozens of top researchers decamped for competitors like Mira Murati’s new Thinking Machines and Meta’s Superintelligence Labs. Fortune

💸 Microsoft’s AI Sales Stumble Exposes Bubble Fears

In a move that sent shockwaves through the AI investing community, Microsoft quietly slashed AI software sales quotas this week after salespeople missed growth targets—with enterprise customers resisting premium pricing without proven productivity gains.

The details:

• Sales teams fell short on 20% to 30% year-over-year growth targets for products like Copilot and Azure AI services.

• Customers are demanding “ROI math, not demos,” according to one anonymous salesperson.

• Microsoft initially denied the report this morning, but shares still fell before recovering slightly as the company pushed back on the characterisation.

• The pressure comes after Microsoft’s $35 billion AI infrastructure spend in Q3 alone.

Why it matters: This could be the canary in the coal mine for the AI bubble. With Bank of America surveys showing 45% of fund managers now flag AI overvaluation as the top risk, Microsoft’s stumble validates concerns that enterprise adoption is plateauing at pilots rather than profits. Big Tech has collectively spent $400 billion on AI capex, but if customers won’t pay premium prices for uncertain gains, the business case for this massive investment starts to crumble. Yahoo Finance

🚗 Nvidia Unveils First Open Reasoning AI for Self-Driving Cars

Nvidia released Alpamayo-R1 at the NeurIPS conference this week, marking a major breakthrough in autonomous vehicle development. It’s the world’s first industry-scale open reasoning vision-language-action model that allows self-driving cars to explain their decision-making process.

The details:

• Alpamayo-R1 integrates chain-of-thought AI reasoning with path planning, enabling vehicles to handle complex scenarios like pedestrian-heavy intersections or double-parked vehicles in bike lanes.

• The model breaks down scenarios and reasons through each step, considering all possible trajectories before choosing the best route—similar to how humans think while driving.

• Based on NVIDIA Cosmos Reason, the open-source model lets researchers customise it for non-commercial use cases and benchmarking.

• It’s now available on GitHub and Hugging Face, with training data included in NVIDIA’s Physical AI Open Datasets.

Why it matters: Previous self-driving models struggled with nuanced situations that required common sense. By giving autonomous vehicles the ability to “think out loud” and explain their reasoning, Alpamayo-R1 represents a critical step toward achieving level 4 autonomy—where vehicles can handle most driving situations without human intervention. The open-source nature democratises access to this technology, potentially accelerating the entire industry’s progress. Nvidia Blog

Quick Hits

• AI Law Firm Raises $2.5M in Days: Logan Brown, 30, quit Big Law firm Cooley and secured a $2.5 million pre-seed term sheet within two weeks for Soxton—an AI-first law firm targeting startups with affordable legal services, bypassing traditional billable hours. The company has already served over 270 early-stage companies. Business Insider

• Luma AI Expands to London: Nvidia-backed video generation startup Luma AI announced plans to hire 200 employees in London by early 2027—about 40% of its workforce—just two weeks after raising a $900 million funding round that valued the company at over $4 billion. CNBC

• AWS Simplifies AI Agent Customization: Amazon Web Services introduced Reinforcement Fine Tuning in Amazon Bedrock at its re:Invent conference, making it dramatically easier for developers to customise AI models using reinforcement learning without extensive ML expertise—potentially reducing inference demands and costs for AI agents. SiliconANGLE

That’s all for today,

See you tomorrow. Tom