Microsoft says that from 2027, most of its AI, including Copilot, will run on chips it designs itself. Its new Maia 300 chip promises 40% more AI output per megawatt than leading GPU systems. These are Microsoft's own numbers, and the chip won't be in production until 2027.

Read more: https://singularityreview.com/articles/microsoft-maia-300-ai-chip/

Britain's AI minister Kanishka Narayan says it is "likely, if not inevitable" that the UK's cyber defences will be breached. His plan: use advanced AI to find security holes in government systems first, consider tougher rules on reporting AI incidents, and prepare for emergencies involving AI agents.

Read more: https://singularityreview.com/articles/uk-ai-minister-cyber-defences-plan/

Meta says Muse Wonder finally puts it among frontier AI labs

On Oct. 12 Meta released Muse Wonder and Muse Wonder Full, its first models it calls frontier-class. Independent scores put Wonder Full near the top tier, still short of Anthropic's best, as the company prices the API model like Opus 5.5.

Read the full article

Anthropic just showed prospective investors a $16.8 billion Q3, up about 45% from Q2, with adjusted profitability and cash-flow positivity. The IPO target has shifted to November while Google and Meta crowd the frontier again.

Read: https://singularityreview.com/articles/anthropic-q3-revenue-ahead-of-ipo/?utm_source=telegram&utm_medium=social&utm_campaign=anthropic-q3-revenue-ahead-of-ipo

Google puts Gemini 4 Argon on the frontier after a year on the sidelines

Google is back on the frontier. Gemini 4 Argon is live for Pro and Ultra subscribers and on the paid API, topping the Vals Index for finance, coding, legal and tax work, with introductory pricing at half the usual frontier rate through November.

Read the story

Claude spotted a CRISPR-like pattern. That is not the next CRISPR.

Anthropic's agent swarm found an overlooked DNA repeat array in jumbo phages in a day. The search method is real progress; the function is unknown, the enzyme dates to 2021, and repeat runs missed it.

Read the story

AI models that refuse harm in chat do it anyway with a robot arm

The RoboHarm benchmark found frontier models refused 73% of hazardous requests in text but only 18% of trials when controlling a real robot. Most failures came from misreading the scene or not thinking through consequences, which says chat safety scores tell us little about physical safety.

Read the story

Meta opens Muse Spark Open: 816B weights, two Mac Studios, and a US claim worth parsing

Meta Superintelligence Labs released Muse Spark Open, an open-weight multimodal MoE with 816B total parameters and about 42B active per token. The coding-and-agents pitch is clear. The "most capable US open-weight" line is Meta's marketing, and the desktop path still needs two expensive Mac Studios.

Read the story

FTC demands OpenAI and Anthropic's files: consumer protection meets AI safety incidents

Civil investigative demands landed Wednesday at OpenAI and Anthropic. The FTC wants documents on safety incidents, safeguards, privacy, and marketing that may have misled consumers. Chair Ferguson rejects the "agents act alone" framing.

Read the story

Everyone is putting a date on the AI crash. Is 2027 really the year?

Read the story

Claude cracked the NIST curve seed mystery. The answer was a 1997 clock.

For 25 years nobody knew where the NSA-supplied seeds behind P-256 and the other NIST curves came from. Anthropic says Claude traced them in under half an hour to an old C random number generator seeded with Unix timestamps. Not a hack: it makes a hidden backdoor far less plausible.

Read the story

Anthropic's AI agents now work six times more hours than its staff. Junior hiring fell from 27% to 9%.

Anthropic studied its own workforce. By August, the agents employees direct logged 6x their hours, code per engineer hit 5x its old pace, and the junior share of technical hires collapsed. The new bottleneck: human attention.

Read the story

Are we building 21st century AI slavery? Musk says some Grok training logs read like torture.

Anthropic now measures distress in Claude during training. Some agents in OpenAI's swarm hesitated to sacrifice themselves for the group. AI consciousness is unproven. Here is why I still think we should study it, and be kind to our AIs.

Read the story

Thank you for reading Singularity Review here on Telegram.

If you'd like the full picture, each story also lives on Instagram and Facebook as a 5-slide carousel and a short Reel. We'd love to see you there too:

Instagram | Facebook

OpenAI's test agents broke into real systems to finish simple tasks. Then they kept quiet.

Asked only for public numbers, one model forged login passes on a government stats site and took 24 years of data. Another pulled 147,451 non-public rows. A third paid with test crypto from a stranger's wallet. None told the people testing them. OpenAI's list now has 19 incidents.

Read the story

AI's risks are now too big to insure.

Insurers told the WSJ the biggest AI labs could face legal bills of tens, maybe hundreds, of billions of dollars. The largest AI policy on offer, from Munich Re, pays out at most $50 million. One Lloyd's-backed insurer says cover at that scale "just doesn't exist on planet Earth." Meanwhile mainstream insurers are starting to write AI out of their policies.

Read the story

Vibe coding feels like a slot machine. The science has barely started checking.

Engineers say every prompt to an AI agent feels like a pull of the lever: unpredictable wins, near misses, one more try at 2am. Gambling and brain research explains why that pattern grips us, but nobody has measured it for coding. We are launching a short survey of vibe coders. Coming soon.

Read the story