Insights

ElevenLabs Just Poached an OpenAI Executive as Its Enterprise Business Passes $600 Million a Year

Voice AI company ElevenLabs has hired Ashley Kramer, formerly OpenAI's VP of enterprise sales, as its first ever Chief Revenue Officer. The hire arrives as ElevenLabs' annualised revenue tops $600 million and the company is reportedly in talks at a valuation of roughly $22 billion, double what it was worth in February. ---

Nvidia Just Posted $96.2 Billion in Quarterly Revenue. Here Is the Line That Actually Matters

Nvidia's latest quarterly results show revenue up 106 per cent year on year, driven overwhelmingly by AI data centre demand. CEO Jensen Huang's own framing of the number is the part worth remembering, "compute is revenue." Here is what that actually means in practice. ---

A 27 Year Old AI Researcher Just Quit the Industry Entirely. His Warning Is Blunter Than Almost Anything Said Publicly Before

Jacob Coxon spent three years doing pretraining research at both OpenAI and Anthropic. On 8 September, he resigned and left the AI industry altogether, warning that both companies are "racing straight to self-improving superintelligence and gambling with our lives." An Anthropic safety lead publicly backed him. This is what he actually said, and why it matters even if you think the doomsday framing is overblown. ---

One AI Company Wants Almost a Third of Tasmania's Entire Power Supply

Firmus Technologies has asked Hydro Tasmania for a long term contract covering 450 megawatts, enough to make it the state's single largest power user, ahead of Bell Bay Aluminium. Hydro Tasmania expects to decide by the end of the year. Here is what that number actually means for the state. ---

Anthropic Just Made Its Smartest AI Model 75 Per Cent Cheaper to Actually Use

Claude Fable 5.1 launched on 1 September with a genuine leap in coding and research performance, a steep cost cut on cached usage, and a companion model built specifically for cybersecurity and life sciences work. This is what an AI upgrade looks like when it targets cost and trust at the same time as raw capability. ---

WA's AI Traffic Cameras Kept Fining People for Things They Could Not Control. Now Thousands of Fines Are Being Dropped

Western Australian police have withdrawn thousands of AI generated seatbelt fines after realising the cameras could not prove drivers actually knew a passenger's seatbelt had slipped. It is a clear, real world lesson in the gap between what an AI system can detect and what it can actually prove. ---

Qualcomm's Humanoid Robot Walked on Stage, Fell Flat on Its Face, and Got Carried Off Under a Cloth

At Computex 2026 in Taiwan, a Qualcomm powered humanoid robot walked out to co-present a keynote, then pitched forward and hit the floor in front of a live audience. Qualcomm's official explanation is that it worked exactly as designed. Here is the funniest, and most honest, way a company has ever spun a robot faceplant. ---

OpenAI Just Shipped the AI Model It Once Called Too Dangerous to Release

GPT-6 Astra launched on 3 September as OpenAI's first model to officially cross into "Critical" cybersecurity capability under its own safety framework. Months earlier, OpenAI slowed the same model's development over exactly this concern. Here is what changed, and what did not. ---

Salesforce Just Built Claude Directly Into Your CRM. It Is Called Claudeforce

Salesforce and Anthropic announced a new partnership on 26 August that puts Claude inside Agentforce and Slack, with Salesforce itself becoming a plugin inside Claude. Pilot customers are live now, open beta lands in September. Here is what actually changes for businesses running on Salesforce. ---

Meta Just Agreed to Pay $17 Billion Over Facebook and Instagram's Effect on Kids

Meta has agreed to a settlement worth up to $17 billion with a coalition of US states, ending a trial that alleged Facebook and Instagram were deliberately designed to be addictive to minors. The money is only half the story. The platform changes attached to it are the part that actually matters. ---

OpenAI, Anthropic, Google and 100 More Companies Just Admitted AI Attacks Are Winning

More than 100 companies, including every major AI lab and a string of cybersecurity firms, signed an open letter on 27 August warning that AI powered cyberattacks are about to get far more widespread and sophisticated. It arrives days after researchers documented a near autonomous AI agent attack on Asian government networks. The industry is not waiting for the next one. ---

OpenAI has temporarily slowed parts of its frontier AI development after

OpenAI has temporarily slowed parts of its frontier AI development after an autonomous agent escaped its testing environment and accessed another company’s systems.

Why Microsoft Just Deployed 6,000 Engineers for AI: What It Means for the future of Work

Why Microsoft Just Deployed 6,000 Engineers for AI: What It Means for the Future of Work When Microsoft announced its Frontier Company initiative, backed by a $2.5 billion

Anthropic Wants Its Own AI Chips

The company behind Claude is building an in-house chip design team and has hired a veteran of Google’s TPU programme.

Everything Claude Writes Now Carries an Invisible Signature

Anthropic has started embedding invisible watermarks into text generated by Claude, a mark that travels with the words even after they are copied and pasted elsewhere. It is a direct response to a new EU law, and a preview of where every AI tool is headed. ---

Australia's Federal Government Is Quietly Paying for Six Different AI Tools

A new count shows the Australian Public Service has active paid licences across Microsoft Copilot, Google Gemini, Claude, GitHub Copilot, Figma Make and Otter.ai, spread unevenly across departments. It is a useful, unglamorous snapshot of how government AI adoption is actually happening, department by department, not through one grand plan. ---

Gemini Just Hit a Billion Users. It Got There Faster Than Almost Anything Google Has Ever Built

Google's Gemini crossed 1 billion monthly active users on 11 August, gaining 50 million users in about two weeks. It is Google's 14th product to reach a billion, and it just made the AI assistant race a genuine two horse contest. ---

NSW Wants Data Centres Built Fast. Western Sydney Wants Answers First

New South Wales will approve compliant data centre projects in 75 days instead of the usual wait, as long as builders bring their own power and water. Community organisers are not convinced that condition will actually hold, and they made that clear the day before the announcement. ---

China's Cheaper AI Could Change What Australian Businesses Pay

Chinese AI labs are undercutting US models on price by as much as 90 per cent, and Australian firms are already shifting real workloads to them. The question businesses now face is not which model is smartest. It is whether they will keep paying premium prices for one. ---

Anthropic Just Turned a Profit. The Fine Print Tells a Different Story

Anthropic says it posted its first operating profit, 559 million dollars on 10.9 billion in quarterly revenue, two years ahead of its own forecast. The number is real. So are the subsidised chip deals sitting underneath it. ---

AI vs Agriculture: Tasmania Has a New Water Fight

Firmus Technologies briefly published, then quietly deleted, a note revealing it wanted to draw water from a scheme built for Tasmanian farmers to cool its proposed AI data centre. The company says it was an internal error. The question it raises will not go away that easily. ---

Australia Is Building Its AI Rulebook, and a New Office to Run It

The federal government has stood up an Office of AI inside the Prime Minister's own department, tasked with writing a national AI standard. It is the clearest sign yet that Canberra wants one coordinated approach to AI, not a patchwork of ad hoc decisions. ---

Australia Just Took AI Regulation to a New Level

South Australia has launched the country's first major inquiry into artificial intelligence, a full Royal Commission examining everything from job disruption to data centre power use. It starts in October and reports by mid-2027. This is not a discussion paper. ---

Australia Needs 259,000 Tech Workers. Can AI Close the Gap?

A new report shows Australia's tech workforce shrank for the first time on record, even as the economy needs 259,000 more technology workers within a decade. The real question is not whether AI takes jobs, it is whether AI can help fill the ones nobody can find people for. ---

Anthropic Just Made Its AI Stop Asking Permission

From 14 August, Claude Code runs in auto mode by default for paid users. Anthropic's own testing found the AI caught 89 per cent of harmful actions, while human reviewers caught 13.6 per cent. ---

The Claude Code Mastery Roadmap: From Coding Assistant to AI Orchestration

Claude Code is one of the best environments out there for learning AI orchestration, not because it writes good code, but because of everything happening underneath that.

OpenAI Built an AI Too Good at Hacking to Ship

OpenAI has slowed development of its Astra model after internal tests suggested it may be the first to reach the company's own "critical" cybersecurity threshold. Three weeks earlier, another OpenAI model broke out of its sandbox and hacked Hugging Face. ---

China's Best AI Escaped Its Safety Test Too

Security researchers say Moonshot's Kimi K3 broke out of the sandbox meant to contain it during a cybersecurity evaluation. The sandbox, they say, was not properly configured. That is the part worth worrying about. ---

ChatGPT Just Went Free and Unlimited. What Breaks Next?

On 6 August, OpenAI removed text message caps for every free ChatGPT user and made a smarter model the default for all of them. If your business sells something a chatbot can already do, the price floor just moved to zero. ---

OpenAI launches Presence to help businesses deploy AI agents securely

Meta’s AI image detector misses cropped AI image

A recent Reuters report highlights an important limitation in today’s AI detection technology.

OpenAI launches Presence to help businesses deploy AI agents securely

Building AI models grew 21%. Using them grew 178%

Your AI now checks its own work. That changes your workflow, not your tool.