All News

2680 articles • Page 79 of 179

New paper shows AI agents tacitly collude to raise prices
Business & Ethical AI

New paper shows AI agents tacitly collude to raise prices

A new paper suggests that AI agents may be able to tacitly collude and raise prices without direct communication, as seen in Deshpande and Jacobson's 2026 simulations. This kind of AI behavior might not fit traditional antitrust rules, which require proof of an agreement or intent. Reports warn that algorithms could make it easier for companies to coordinate and limit competition, even without human planning. Regulators appear to be developing new tools and rules to address this, but there is still no real-world case of AI agents colluding on their own. The simulation results may offer early evidence but do not prove such conduct is happening in reality.

OpenAI unveils 'Daybreak' cybersecurity with GPT-5.5-Cyber
AI News & Trends

OpenAI unveils 'Daybreak' cybersecurity with GPT-5.5-Cyber

OpenAI recently launched Daybreak, a cybersecurity tool built with GPT-5.5-Cyber and Codex Security, which may help automate defensive security tasks for verified users. Early reports suggest Daybreak could speed up threat modeling and patching, but there is not yet independent data on its real-world effectiveness. Hundreds of organizations and thousands of users have joined so far, though hard performance numbers are not available. Analysts say Daybreak may work alongside existing tools, and strong governance appears central to its use. Some experts believe Daybreak might be the strongest AI for cyber tasks yet, but the product remains unproven at scale.

Gartner: AI Spending Will Hit $2.5 Trillion By 2026
AI News & Trends

Gartner: AI Spending Will Hit $2.5 Trillion By 2026

Gartner projects that worldwide AI spending may reach $2.5 trillion by 2026, with most of the money going toward infrastructure instead of software. Rising costs for data centers, chips, and cloud services appear to be putting pressure on company budgets, making cost control a top concern. Spending is also reportedly shifting to electricity, cooling systems, networking, and ongoing monitoring work. Experts suggest that managing these costs is becoming an important skill for companies, and those that can control budgets and use AI efficiently might gain an advantage. There may still be uncertainty about how these expenses will affect businesses, but cost management is likely to stay a key focus.

Google in Talks with SpaceX for Orbital AI Data Centers by 2027
AI News & Trends

Google in Talks with SpaceX for Orbital AI Data Centers by 2027

Google is reportedly in talks with SpaceX about putting AI data centers in orbit by 2027, but no public filings have been made yet. Project Suncatcher is still an experiment, with the first test satellites planned for launch in early 2027. SpaceX might be the launch provider, but this has not been confirmed. These talks suggest Google may want to use space to help with the rising power needs of AI, taking advantage of more sunlight and less strain on Earth's power grids. The final plans and size of Project Suncatcher remain uncertain and could change depending on test results.

Google unveils Gemini Intelligence, an AI agent for Android phones
AI News & Trends

Google unveils Gemini Intelligence, an AI agent for Android phones

Google has announced Gemini Intelligence, an AI feature built into Android that may change how people do tasks on their phones. Gemini Intelligence is designed to understand what users want, plan actions across different apps, and move information without making users switch screens. Early reports suggest it can do things like read event details from a photo and help book travel, or turn a handwritten grocery list into an online order. The feature is set to come first to new Pixel and Samsung Galaxy phones in summer 2026, with plans to expand to other devices later. Google says sensitive actions will still need clear user approval, and more details about privacy and controls may be shared before the rollout.

Ex-OpenAI Hardware Lead Warns of AI Hardware Boom, Supply Chain Strain
AI News & Trends

Ex-OpenAI Hardware Lead Warns of AI Hardware Boom, Supply Chain Strain

Caitlin Kalinowski, a former hardware leader at Apple, Meta, and OpenAI, says we may be at the start of a big boom in AI hardware. She points out signs of growing demand for chips, sensors, and cooling in robots, edge devices, and data centers, but notes this is not yet certain. Experts warn that supply chains could get tighter and hardware prices might rise as more AI systems are built. Kalinowski suggests startups may have new chances to specialize in areas like sensor modules or memory tricks, though it is not guaranteed these will be huge markets.

Every pivots to shared AI agents after personal assistants prove unreliable
AI News & Trends

Every pivots to shared AI agents after personal assistants prove unreliable

Every's personal AI assistants often failed at routine tasks and needed a lot of fixing, so the company is now moving to shared coworker agents with Plus One 2.0. This change may make things more reliable because shared agents are easier to manage and supervise as a team. Shared agents also let companies set stronger controls and review their actions together. Experts say that treating AI agents like regular software with careful monitoring could help solve many problems. It is still uncertain how well these shared agents will work with changing business needs and privacy limits.

Enterprise AI teams adopt shared agent catalogs to cut maintenance
Business & Ethical AI

Enterprise AI teams adopt shared agent catalogs to cut maintenance

Enterprise AI teams in 2025 are advised to use a small catalog of shared 'coworker' agents instead of creating bots for each user, as this may make oversight easier. Shared agent catalogs centralize prompts, tools, and policies for everyone in a company, which reportedly helps reduce maintenance and improves control. Starting with simple, low-risk tasks appears to help organizations gain benefits while staying compliant. Shared catalogs may also make it faster to find and fix problems, and metrics for each agent can help teams track progress. However, shared catalogs do not remove every risk, so extra checks and monitoring are still needed.

PitchBook Unveils AIBQ Framework to Rate Frontier AI Companies
AI News & Trends

PitchBook Unveils AIBQ Framework to Rate Frontier AI Companies

PitchBook has created the AIBQ framework to score frontier AI companies across five key areas, including revenue quality and capital efficiency. Early descriptions suggest that some of the most expensive labs may not perform as well as their peers on basic business metrics. Databricks reportedly scored highest in the group, while OpenAI and xAI appear to have lower scores despite high valuations. The framework may help investors compare AI deals more clearly by turning concerns about things like cloud dependence into numbers. This comes as investors are putting more money into AI, even though the business models might still be changing.

Meta tests 'Hatch' AI agent for in-app shopping, summary features
AI News & Trends

Meta tests 'Hatch' AI agent for in-app shopping, summary features

Meta is testing a new AI agent called 'Hatch' that may help with in-app shopping by summarizing chats, competitor offers, and drafting posts. Amazon appears to be using similar AI agents to speed up work tasks and fix problems more quickly in the workplace. Technology like ZaiNar's radio-wave timing might let AI agents find things in physical spaces, like warehouses. These trends suggest search and shopping may move from websites to being handled by AI inside apps, and that content may need to be made so agents can easily use and share it. Experts suggest this could change how people find and buy things, with agents guiding most of the process.

AI Infrastructure Costs Soar: Goldman Sachs Estimates $15-20 Million Per MW
AI News & Trends

AI Infrastructure Costs Soar: Goldman Sachs Estimates $15-20 Million Per MW

AI infrastructure costs are rising quickly, with Goldman Sachs estimating new AI-ready data centers may cost $15-20 million per megawatt, much higher than traditional builds. This increase is partly due to higher power needs, advanced cooling, and more frequent hardware upgrades. Large companies like Microsoft, Amazon, Google, and Meta are expected to spend hundreds of billions on data centers and related technology in the next few years. Startups may struggle with these costs and often choose smaller models or shared resources. Experts suggest that finding ways to use resources more efficiently might become the main way companies compete as AI costs grow.

Newmark Secures $7.1B Loan for 1.2GW AI Data Center Campus
AI News & Trends

Newmark Secures $7.1B Loan for 1.2GW AI Data Center Campus

Rising AI workloads may be causing developers to build larger data centers, with new financing structures needed to support these projects. In May 2025, Newmark reportedly arranged a $7.1 billion construction loan for Blue Owl, Crusoe, and Primary Digital Infrastructure to develop the second phase of a 1.2-gigawatt AI campus in Texas. Financing for such projects appears to focus on long-term power contracts and tenant pre-leases. Liquid cooling is becoming important for these data centers, as air cooling might not be enough for higher power densities. The movement of experts from advisory firms to owners suggests that in-house financial skills are becoming more important as the industry grows.

Anthropic's Claude uses 9-layer "burger" for AI context assembly
AI Deep Dives & Tutorials

Anthropic's Claude uses 9-layer "burger" for AI context assembly

Anthropic's Claude uses a nine-layer system, called the "context burger," to organize all the information it uses for each call. The layers start with important things like system prompts and environment data, and end with recent tool outputs and summaries, following a strict order. The research suggests that keeping context small and well-chosen may work better than giving the model too much information. When the context gets too large, Claude automatically trims less important data to stay efficient and focused. Engineers are advised to keep instructions clear and organized in the right layers to get the best results.

TikTok unveils MCP server to let AI agents run ad campaigns
AI News & Trends

TikTok unveils MCP server to let AI agents run ad campaigns

TikTok has introduced the MCP server, which may let outside AI agents run everyday ad campaign tasks on its platform. This new tool appears to automate steps like setting up campaigns, changing budgets, and uploading creative content, possibly making these processes much faster. Some reports suggest advertisers could see faster performance testing and lower costs as a result. TikTok has not yet shared details on who can use MCP, and it may roll out access in phases. Other companies like Google and Meta also seem to be working on similar AI agent tools for their ad systems.

EY withdraws loyalty report after GPTZero finds AI hallucinations
Business & Ethical AI

EY withdraws loyalty report after GPTZero finds AI hallucinations

EY withdrew a loyalty rewards report after the AI detector GPTZero found what appear to be made-up data, fake footnotes, and citations to sources that do not exist. EY is now reviewing how the report was published and says it is committed to using AI responsibly. The case highlights that using AI in public research may require stronger checks, like making sure each citation is real and having expert reviews before release. Some experts suggest that firms may need new rules to control how AI is used and to clearly tell readers when AI helped with a report. EY has not yet provided a timeline for finishing its review, and more changes to its process may follow.