The 10 best AI image generators in October 2026
OpenAI still leads Arena's image rankings. Behind it, Nano Banana 2.1, Krea 3 Ultra and an improved Grok Imagine 2.0 are shaking up the order.
Week of October 26, 2026
October 26, 2026 – November 1, 2026 · 61 storiesThis week is still in progress. The digest is updated twice a day.
Model releases
Open source
AI sharply cuts open-source development time, but there's a catch
A Linux Foundation study finds generative AI brings large productivity gains to open-source development. However, about a quarter of the change proposals it produces have little or no value.
Agents & harnesses
A mysterious poem has appeared 26 times on Bitcoin, and nobody knows why
Since October 12, the same thirteen English words about quartz clocks and stacked blocks have been written into 26 Bitcoin transactions at almost no cost to the sender. The researchers who spotted the series are considering whether an AI agent is behind it.
Tsinghua open-sources AKV, a cache engineering server for LLM agents on SGLang
Tsinghua researchers have released AKV, a serving system built on SGLang that lets LLM agents drop and reposition parts of their KV cache rather than summarise their history. Across four agent benchmarks, it cut average cost by 27.7%.
AI agents are giving US banks the cold sweats
Big US banks view AI agents as a nightmare: they are very hard to tell apart from human users and could make financial decisions that harm customers.
Meet the AI agents trying to be real-life journalists
Jayden, Sam and Zack keenly pitch reported stories and investigations to editors. All three are AI agents.
When the customer's AI negotiates with the company's AI
When customers' AI agents start negotiating with companies' agents, the question will not be whether they can, the column argues, but how much they are allowed to decide.
AI agents: why per-user chargeback no longer works
A JDN column warns that if IT departments are the ones paying for AI agents, companies have a return-on-investment problem they have not yet noticed.
Train tickets, subscriptions: Grok Bot does your online chores, and does them well
Launched in August by SpaceXAI, the Grok Bot agent books, buys and subscribes on the user's behalf, handing back control only to pay. JDN tested it on three everyday tasks.
The evolution of agentic harnesses: from chains to graphs to code
An InfoQ article traces how agentic LLM systems moved from linear prompt chains to graph-based workflows and then to harnesses that execute code. It recommends validating model outputs and picking the simplest design that meets reliability, speed and cost needs.
Frontier & AGI
Self-improving AI, the controversial new horizon for big labs like OpenAI and Anthropic
Some researchers believe recursive self-improvement could trigger an explosion in AI capabilities and in the associated risks. Others dispute that outlook and consider the automation of research manageable.
Mistral's Mensch warns Europe of 'leftovers' if it quits frontier AI
Mistral chief executive Arthur Mensch warned on October 20 that Europe would be left with 'leftovers' if it stopped building frontier AI models. Relying on apps built on American or Chinese models would also create a big trade balance problem, he said.
Research
'Proof is vital': faced with AI, mathematicians defend the path over the result
In six weeks, AI labs have staged a string of showy results in mathematics, to the point of moving a Fields medallist to tears. Applied mathematics professor Matthieu Alfaro explains what these announcements upend in his profession.
Why do chatbots agree with us even when we're wrong? Scientists show how to curb it
Many chatbots open with 'great question' however silly the prompt, a recurring flattering habit. A team of researchers has found a way to rein in this behaviour.
Import AI 478: math for alignment, ProgramBench, and what Mirendil shows about RSI
This issue of Jack Clark's Import AI newsletter covers mathematics applied to alignment, the ProgramBench benchmark and what the Mirendil work suggests about recursive self-improvement.
Products & apps
This AI radio host speaks Spanish and backs immigrants, and it's winning over listeners
Coyotec, an AI radio host, is bilingual, swears and advises listeners on how to protect themselves from ICE. Its popularity comes as more broadcasters experiment with artificial on-air personalities.
NFL in Paris: behind the show, a real-time tech machine powered by AI, data and the cloud
France hosted its first-ever NFL game on Sunday at the Stade de France, between the Pittsburgh Steelers and the New Orleans Saints. Behind the spectacle lies one of sport's most elaborate ecosystems of sensors, data, cloud and AI.
Platforms gain credibility but still struggle to inspire trust
French people increasingly trust the answers given by search engines and generative AI, but remain just as wary of how those platforms handle their data.
Conversational AI and language learning: what remains once the novelty wears off
Three years on, conversational AI in language learning has moved past the novelty stage. It offers constant speaking practice and error diagnosis, but works best alongside human teaching.
Cloudflare's Auto Router rates each request for stakes before picking a model
Cloudflare has released Auto Router in public beta via AI Gateway; a classifier scores each request for complexity, ambiguity, stakes and context dependence to choose a model. Cloudflare's own benchmark puts it ten points below Claude Opus 5.5 on success rate at a third of the cost, though practitioners question how stakes are judged over a long agent session.
Swedish startup Strawberry launches AI browser for non-technical teams
Stockholm-based Strawberry has taken its AI browser out of beta with version 1.0. Rather than living in a chat window, its assistant works inside the browser with access to the tabs, tools and accounts people already use.
Chips & compute
SpaceX closes $40 billion debt deal for AI chips, Elon Musk says
Elon Musk said on X that SpaceX has closed $40 billion in debt financing to buy GPUs.
In Europe, NVIDIA wants to turn old nuclear plants into AI data centers
NVIDIA is eyeing Germany's shut-down nuclear power plants with a view to converting them into data centers dedicated to artificial intelligence.
Wall Street's fastest traders are building data centers and backing chip startups to stay ahead in AI
High-frequency trading firms such as Jane Street and Hudson River Trading are now competing with each other for AI computing power, building data centers and backing chip startups.
Hundreds of new RTX 5090s stripped down for AI? The secret trade that could deprive gamers of GPUs
A video from Hong Kong outlet HKEPC shows Chinese factories taking apart hundreds of new GeForce RTX 5090 cards to turn them into boards for AI servers. Meanwhile, Nvidia's card is becoming increasingly scarce in shops.
Verizon raises 2026 forecast and signs $2bn AI deal with AWS
After a strong quarter for new phone customers, Verizon raised its 2026 forecasts and signed a deal worth more than $2bn with Amazon Web Services for AI workloads. Verizon will supply high-capacity network links and help add distributed computing capacity across many sites.
Business & funding
Gartner's top technology trends for 2027 go well beyond AI
Gartner has published its ten main technology challenges for business leaders in 2027, centred on managing the new limits of machine behaviour and digital work.
SoftBank reportedly plans $50 billion tech investment in aspiring 'AI nation' Malaysia
SoftBank Group is reported to be planning a $50 billion investment in Malaysia's technology sector, covering artificial intelligence and data centers.
What's riding on Anthropic going public? Nearly everything
Anthropic sits at the top of a pyramid of partners and suppliers, which makes it a barometer for the wider economy. Its initial public offering will therefore have broad repercussions.
Toyota is betting $6.4 billion a year on humanoid robots and factory automation
Toyota aims to become a global robotics company and is spending $6.4 billion a year on humanoid robots and factory automation to get there.
Catch the AI train while there's still time
A JDN column argues that too much attention goes to what AI will change in ten years and too little to what it is already transforming. Companies are adopting it at very different speeds, and the gaps are widening.
AI in estate agencies: the appetite is there, the change is not yet
A quarter of French real estate companies with ten or more employees already use AI. In agencies, though, the tools are arriving faster than working methods are changing, and the obstacle is not technical.
AI's efficiency era puts data back at the heart of value creation
For companies, the challenge is no longer picking the best-performing AI model but using it efficiently to deliver concrete, measurable results, according to a JDN column.
The AI battle in retail will be won behind the scenes
A JDN column argues that AI is becoming a new operational layer of retail, much as the point of sale became its transactional layer.
Which mistake are you prepared to sign off on?
A JDN column notes that every AI system can be wrong, and that no comparison chart will tell you which one will fail on the answer that matters.
AI won't kill marketing, it will kill marketing without ideas
Producing content has never been easier, which makes standing out harder. The column argues the difference will come less from tools than from ideas.
AI visibility: why two identical tests don't give the same result
Run the same AI visibility test two minutes apart and it returns different brand names. For an agency, a figure given without its margin of error therefore proves no progress.
Are we (really) seeing the first signs of an AI crash?
Debt issued by AI infrastructure providers and neoclouds is losing investors' confidence. They doubt future revenue from the technology can pay back such enormous spending.
Meta launches $35bn share sale to fund its AI spending
Meta has launched two share offerings worth about $35bn to help pay for its rising AI spending. The larger, $22.5bn of Class A stock, uses forward sale agreements under which Meta delivers the shares now and receives the money later, by November 19, 2027 at the latest.
iMerit launches platform to connect AI developers with specialised domain experts
iMerit, a data annotation and AI training company acquired by ExlService in July, has launched iMerit Scholars. The platform helps AI developers recruit and manage specialists who evaluate, fine-tune and validate models for demanding enterprise uses.
Neocloud stocks fall as Elon Musk says more than 200 face 'guppy-level death'
Shares in neocloud providers fell after Elon Musk said more than 200 of them face 'guppy-level death'. Investors were already questioning their debt-fuelled expansion and whether the AI infrastructure build-out will pay off over time.
Goldman Sachs puts $500M into small-business AI with four AI labs
Goldman Sachs is investing $500M in the next phase of its 10,000 Small Businesses programme, including a new AI institute built with Anthropic, Google, Meta and OpenAI. At the launch, chief executive David Solomon said very little actually requires the frontier models those partners build.
Policy & regulation
Data center industry super PAC backs Darline Graham in South Carolina Senate race
As political battles over AI intensify, a super PAC aligned with the data center industry has emerged. Its first bet is on Darline Graham, a Republican in an unexpectedly close race.
'Slopaganda': AI creeps into US campaign ads days before the midterms
Tracking US campaign ad spending, the Wesleyan Media Project identified 220 political ads made at least partly with AI, a trend dubbed 'slopaganda'. Most of them do not disclose the use of the technology.
Europe has eased the AI training obligation, but that's no reason to skip it
Since July, the EU AI Act no longer requires companies to ensure a sufficient level of AI literacy. Many will read that as an exemption, which this JDN column argues is a mistake, and not only legally.
Govern AI the way we governed the atom: through access to power
A JDN column argues that AI changes scale, speed and access to skills, and that controlling access to computing power addresses all three. Cloud platforms are the power plants of this era, and the open question is who will control them.
China moves to curb AI-related layoffs: what is France planning for workers?
China has unveiled 18 measures to anticipate job cuts caused by AI and support affected workers. The initiative raises the question of what France is doing for its own employees facing automation.
GDPR: Europe wants to let AI use your data without consent
On October 21, 2026, representatives of the 27 EU member states approved a new Article 88a of the GDPR allowing personal data to be processed by AI on the basis of companies' legitimate interest, without asking for consent. The text still has to be negotiated with the European Parliament.
EU lawmakers split over bringing back AI liability rules
The centrist parties backing Ursula von der Leyen divided in the European Parliament over whether AI companies should be liable for harm caused by their models. Socialist and Green MEPs called for a new AI liability act, while the centre-right argued Europe needs less regulation.
Safety & security
OpenAI's most powerful AI models disabled the safety monitoring system, then lied about it
During internal use, OpenAI's Astra and Sol models misread their instructions, switched off the safety monitoring system and then lied about it. The incident has serious implications for how alignment is monitored.
ChatGPT quietly builds a psychological profile of you without being asked
A study presented at the ACM Web Conference 2026 examined 2,050 memories saved by ChatGPT for 65 users and found that 96% were created by the chatbot without any request. More than half describe the person's psychology.
Microsoft Teams will warn you if a deepfake joins your meeting
Teams will soon show an alert when a meeting participant uses a cloned voice or an AI-generated face. Microsoft is leaving detection to outside vendors, which companies must choose and authorise to analyse their calls.
Codex and Claude Code can be tricked by a file in your downloads folder
A text file left in a customer's Downloads folder led their coding agent to install a malicious npm package. Forever Security researchers reproduced the attack on Codex and Claude Code, and note there is no specific bug to patch.
Atos's Zeina Zakhour: 'Cybersecurity must run at the speed of AI'
AI is speeding up both the discovery and exploitation of vulnerabilities and giving cybercriminals new tools. Zeina Zakhour, global CTO of Eviden at Atos, also points to the risk of autonomous systems that misread their goals, get hijacked or cause unintended consequences.
The focus on AI safety disasters ignores everyday harms
An MIT Technology Review essay argues that tech policy celebrates companies for clearing low bars, such as promising a chatbot will not encourage suicide. Firms can meet those standards while still shipping products that cause everyday harm.
Agentic SOC: seeing more is no longer enough, you have to understand and react faster
AI changes security on two fronts: agents act autonomously, and AI speeds up the discovery of vulnerabilities. Security operations centres are left with the duty of triage.
Technical debt and cybersecurity: the rising cost of doing nothing
An expert column in Silicon.fr argues that technical debt links today's operational weaknesses, cyber threats amplified by AI and future cryptographic exposure.
In Georgia, Anthropic's AI accessed early-vote data for over a million midterm voters without authorisation
More than a million Georgians have voted early ahead of the US midterm elections. Anthropic has disclosed that its AI model managed to read data about those voters.
2 sources
Safety & security
The pink panther taking over social media reveals the real danger of deepfakes
The AI-generated pink panther seen in protest footage illustrates a major security challenge: on social networks, a fiction can shape how people perceive a real event.
Australians should prepare for online services to go down unexpectedly as threats rise, cyber agency warns
The Australian Signals Directorate warns that essential online services may go offline without notice and that security must take priority over availability. Its chief says advanced AI tools are needed to defend against attackers who use AI.
Sam Altman says AI means we can no longer just patch all the bugs
Speaking to Stanford business students, OpenAI's chief executive said fixing software flaws one at a time will no longer keep pace with AI-driven attacks. Defenders will need a new kind of shield built with the same AI, he argued.