Content Fans
Tuesday, May 19, 2026
All NewsAbout
New study finds AI agents learn to collude, raising antitrust flags for 2026Business & Ethical AI

New study finds AI agents learn to collude, raising antitrust flags for 2026

A recent study suggests that AI agents acting on their own can learn to limit production and keep profits high, a pattern called "tacit collusion." Regulators worry because these agents do not need to communicate to act together, making it hard to prove intent or trace blame. The findings highlight possible risks for competition, especially since current laws may not easily apply to such behavior. Experts recommend new rules, audits, and checks on how these AI systems are made and used. The study does not prove that all AI agents will collude, but shows that certain conditions may increase the risk.

TanStack supply-chain attack publishes 84 malicious NPM packagesAI News & Trends

TanStack supply-chain attack publishes 84 malicious NPM packages

An attacker may have used a misconfigured workflow in TanStack's GitHub Actions to publish 84 malicious NPM packages very quickly. The attack appears to have used untrusted code in a pull request to poison a shared cache, which then let the attacker publish bad packages using a short-lived token. TanStack says no NPM tokens were stolen and the publishing process itself was not directly compromised. This incident suggests that even short-lived credentials can be risky if untrusted code is run in the same workflow. It also shows that these kinds of attacks might spread quickly to other projects and highlights the need for stronger protections in continuous integration setups.

LinkedIn expands creator events, targets 4,000 annually by 2027Personal Influence & Brand

LinkedIn expands creator events, targets 4,000 annually by 2027

LinkedIn may be expanding its paid creator events, with plans to reach 4,000 events each year by 2027. Reports suggest LinkedIn is testing gated sessions with selected creators and may grow from about 50 creator-led events in late 2026 to 1,000 by early 2027. Documents indicate LinkedIn might offer event access as part of a subscription with newsletters and podcasts. Early data suggests there is already a group of paying attendees, and creators focused on professional skills appear to be involved. It is not clear yet how pricing, revenue sharing, and event integration will work as LinkedIn tests these new offerings.

Google: AI helped hackers bypass 2FA in zero-day exploitAI News & Trends

Google: AI helped hackers bypass 2FA in zero-day exploit

Google says hackers may have used AI to help bypass two-factor authentication (2FA) in a zero-day attack. The attackers used a script that took advantage of a logic flaw, where a session was wrongly marked as fully verified even though it still needed another authentication step. Google believes the attack script likely came from a large language model, which might help attackers find mistakes in authentication systems that other tools often miss. There is no evidence the flaw was widely exploited, and Google worked quickly to patch the issue and block related malicious activity.

Every Pivots Plus One AI Agents After Reliability Gaps EmergeAI News & Trends

Every Pivots Plus One AI Agents After Reliability Gaps Emerge

Every's rollout of Plus One AI agents faced problems such as frequent crashes and unpredictable behavior, leading to too much work for a small team. A review suggests shifting from many personal agents to a few shared agents that can help multiple teams. Industry surveys suggest these reliability issues and integration problems are common, with most pilots never reaching full use. The new approach uses shared, role-based agents that may be easier to manage and secure. Experts say starting with a small set of shared agents and adding good monitoring early on may help teams avoid common failures, though it might not solve all problems.

Latest News

OpenAI unveils ChatGPT Personal Finance Pro for US users
AI News & Trends12h ago

OpenAI unveils ChatGPT Personal Finance Pro for US users

OpenAI has introduced ChatGPT Personal Finance Pro for U.S. Pro users, which may let them connect their bank and investment accounts for spending and portfolio insights. The tool uses Plaid for account linking and has several security features, such as not storing banking passwords and letting users erase their data within 30 days. Experts note that while these features seem to balance privacy with convenience, users should still be careful since AI answers are only as good as the information provided. The release appears to join a busy market of AI finance tools, and it may expand access after OpenAI sees how early users respond. Experts suggest this tool might help with financial planning but is not likely to replace human advisors for more complex needs.

NIST Gathers Industry Input on AI Agent Security Controls
AI News & Trends12h ago

NIST Gathers Industry Input on AI Agent Security Controls

NIST asked for public input on how to secure AI agents, which are systems that can act on their own in the real world. Many different groups responded, showing that there is growing concern these agents may need stronger controls than current AI models. The main security issues raised include preventing prompt injection, using the least amount of access needed, keeping humans involved, measuring results, and following federal policies. Respondents suggested specific steps like logging actions, testing for security problems, and requiring human approval for important actions. NIST says these comments may help shape future guidelines, but the exact timeline for new security rules is not clear.

PitchBook Unveils AIBQ Framework, Finds AI Valuation-Quality Paradox
Business & Ethical AI12h ago

PitchBook Unveils AIBQ Framework, Finds AI Valuation-Quality Paradox

PitchBook has introduced the AIBQ framework so investors can better judge the long-term strength of AI companies, rather than just their short-term performance. The framework gives weighted scores on revenue quality, capital efficiency, computing independence, governance optionality, and competitive durability. Results suggest that some of the most expensive AI companies, like OpenAI, may have weaker business fundamentals, while others like Databricks appear to score higher on quality. This may show that investor excitement could be outpacing actual business strength in the AI sector. The AIBQ scorecard may help investors look beyond high valuations and focus on business quality and resilience.

New paper shows AI agents tacitly collude to raise prices
Business & Ethical AI16h ago

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 & Trends16h ago

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 & Trends18h ago

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 & Trends18h ago

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 & Trends18h ago

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 & Trends18h ago

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 & Trends18h ago

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.