Friday, June 5, 2026

The token bill comes due: Inside the industry scramble to manage AI’s runaway costs

The AI industry is grappling with soaring costs as companies like Uber have already exhausted their entire 2026 AI budgets by April. Microsoft has revoked Claude Code licenses for developers, and a Priceline employee noted a staggering 4-5x increase in contract renewal costs for AI tools. Despite a decrease in per-token prices, demand for AI has led to skyrocketing consumption, prompting firms to seek better visibility and control over their spending. In response, the Linux Foundation announced the formation of the Tokenomics Foundation, aiming to establish standards for tracking and managing AI token expenses.

The latest AI news we announced in May 2026

At Google I/O 2026, the company introduced the Gemini 3.5 model and Gemini Omni, marking the beginning of the 'agentic' era in AI. Gemini 3.5 enhances complex, multi-step workflows, while Omni allows the generation of high-quality videos from diverse inputs like text and images. The updates also include the new Google Health app for comprehensive health management and the Fitbit Air, a compact tracker featuring advanced health metrics. These innovations aim to make AI more proactive and integrated into daily life.

NSA said to be readying Anthropic’s Mythos for use in cyber operations

The National Security Agency (NSA) is reportedly preparing to utilize Anthropic's AI model, Mythos, for cyber operations, according to a report by the Financial Times. Anthropic has dispatched around six engineers to the NSA to assist in applying Mythos, although it remains unclear if they are actively involved in hacking operations. This development follows an earlier Axios report indicating that the NSA was using Mythos despite a federal ban imposed by the Department of Defense, which designated Anthropic as a 'supply chain risk' due to its refusal to allow government access for domestic surveillance and autonomous weaponry.

This AI startup says it can tell if a script will make a hit film

AI startup Quilty offers a tool that claims to predict a film's box office success based on script analysis, providing a score from 0 to 100. Founded by film producers Simon Horsman and Daniel Wood, Quilty combines various AI tools to evaluate narrative quality, commercial viability, and audience resonance. However, initial tests have raised skepticism, as Quilty inaccurately predicted that the script for the box office flop Christy would outperform the Oscar-winning Sinners. The service aims to assist up-and-coming filmmakers by providing detailed reports on their scripts, but its reliability remains unproven.

Amazon unveils latest warehouse robot as tech giants continue AI layoffs

Amazon has introduced its next-generation warehouse robot, Proteus, which can respond to natural language commands, enhancing the interaction between workers and technology. Launched at the Delivering the Future event in London, this autonomous mobile robot is designed to transport items in fulfillment centers. The original Proteus was deployed in 2022 and is currently used in 25 U.S. centers, with plans to roll out the new version in Europe by mid-2027. This move comes as Amazon continues to lay off thousands of corporate employees while investing heavily in AI-driven automation.

AI is designing OpenAI's next model in a sign of 'superintelligence': SoftBank's Masayoshi Son to CNBC

SoftBank CEO Masayoshi Son announced that OpenAI's next model is being designed by an AI, marking a significant step towards what he describes as 'superintelligence.' In an interview with CNBC, Son revealed that he had spoken to OpenAI CEO Sam Altman, who confirmed that an AI model is responsible for designing future iterations. Son, who uses OpenAI's ChatGPT extensively, predicts that AI will soon outperform humans in 70% to 80% of subjects, potentially being ten times smarter than the average person in those areas. His previous timeline for artificial superintelligence, initially set at ten years, has now been revised to just two years.

AirTrunk commits $30B to build 5GW of AI data centers in India

AirTrunk, backed by Blackstone, has announced a $30 billion investment in India by 2030 to develop 5 gigawatts of new data center capacity, marking one of the largest commitments to the country's digital infrastructure. This follows AirTrunk's entry into India earlier this year via the acquisition of Lumina CloudInfra. The Indian government is actively promoting AI infrastructure investment, offering tax exemptions to foreign cloud providers through 2047 for workloads run from Indian data centers. AirTrunk's plans include a significant 3GW facility in Maharashtra, supported by a letter of intent for land allotment from the state government.

Has Microsoft Lost Its Mojo (Again)?

At Microsoft's annual Build developer conference, CEO Satya Nadella highlighted the company's commitment to agentic AI, despite facing significant challenges. While competitors have seen stock prices soar, Microsoft’s shares have fallen this year, and its Copilot AI tools have not gained expected traction. GitHub, a subsidiary, has suffered unprecedented downtimes, leading to user complaints and even departures, raising concerns about Microsoft's standing in the developer community. The company is now under pressure to regain the trust of developers, a critical demographic for its future success.

Where investors may find the next 'big wave' for AI trade

Tim Urbanowicz, chief investment strategist at Innovator from Goldman Sachs Asset Management, is advocating for investors to focus on emerging markets, particularly Taiwan and South Korea, as the next significant opportunity in artificial intelligence. He noted that these regions are integral to the iShares MSCI Emerging Markets ETF, which has risen 26% this year. Urbanowicz believes that valuations in these markets have not escalated as much as in the U.S., suggesting substantial potential for growth in AI-related investments.

How OpenAI Built a Secure Windows Sandbox for Codex Agents

OpenAI has introduced a custom Windows sandbox architecture for its Codex coding agent, addressing the limitations of existing Windows isolation mechanisms. The new sandbox allows Codex to run locally on developer machines while restricting its access to sensitive areas of the system. OpenAI's solution includes an unelevated sandbox that uses Windows security identifiers and access control lists, and an elevated sandbox that creates dedicated local accounts for executing commands with restricted tokens. This approach aims to balance security with the need for seamless developer workflows.

The Tech Download: Anthropic’s IPO sets up first big test of AI boom valuations

Anthropic has taken a significant step towards an initial public offering (IPO), positioning itself to potentially outpace rival OpenAI in the public markets. The company recently filed a confidential prospectus with the SEC, boasting a valuation of $965 billion and a revenue run rate of $47 billion. This IPO is set to be a critical test for the appetite for pure-play frontier AI companies, as experts highlight that gross margin will be a key metric in assessing the sustainability of such high valuations. With strong competition from well-funded players like Google and Meta, Anthropic's market lead could be challenged as it navigates this pivotal moment.

Microsoft trained its MAI models on unlicensed web data despite promising "enterprise grade, clean and commercially licensed data"

Microsoft has acknowledged that its MAI models were partially trained on unlicensed web data, despite earlier assertions that the training utilized only 'enterprise grade, clean and commercially licensed data.' The revelation comes from a technical paper that cites sources like Common Crawl. Microsoft claims to use a proprietary crawler that respects the Robots Exclusion Protocol, placing the onus on site owners to protect their content. This practice aligns with industry norms among AI companies, yet raises questions about the validity of Microsoft's previous claims regarding data cleanliness.

What to Expect From Apple’s AI, Siri and iOS 27 Launch at WWDC 2026

Apple Inc. is set to announce a comprehensive artificial intelligence strategy at its Worldwide Developers Conference on Monday, aiming to rectify the shortcomings of its previous AI rollout two years ago. The keynote will introduce a significantly overhauled version of the Siri digital assistant, alongside a range of new AI features and operating system updates for the iPhone, iPad, Mac, smartwatch, TV set-top box, and Vision Pro headset. The upcoming software will be designated as iOS 27, iPadOS 27, macOS 27, tvOS 27, watchOS 27, and visionOS 27, reflecting a unified branding across its product ecosystem.

Apex Space Doubles Valuation to $2.3 Billion in Funding Round

Apex Space has successfully raised over $200 million in its latest funding round, resulting in a valuation of $2.3 billion, which marks a doubling from previous estimates. The Los Angeles-based startup specializes in manufacturing satellite buses that manage operations such as power generation and communication. These platforms are designed to accommodate various equipment, including cameras and defense technologies, and are pivotal for both commercial firms and government projects, notably including the development of President Trump's Golden Dome missile defense shield.

I didn’t need a Fitbit Air, but it’s been a great Pixel Watch companion

The Fitbit Air has emerged as a valuable companion for Google Pixel Watch users, despite overlapping functionalities in health tracking. While both devices monitor heart rate, workouts, and sleep, the Fitbit Air's screen-less design offers a lightweight alternative that enhances comfort during sleep and daily activities. Users can enjoy the convenience of switching from the Pixel Watch to the Fitbit Air at night, ensuring the watch is fully charged for daytime use. This combination allows for effective tracking without the distraction of notifications, making the Fitbit Air a worthwhile addition for those already invested in the Google Health ecosystem.

Meta Ray-Ban glasses may soon identify faces using AI

Meta's Ray-Ban smart glasses may soon incorporate an AI-driven face recognition feature known as NameTag, which identifies users from images and videos captured by the glasses. This development was revealed in a WIRED report, highlighting that NameTag has been under construction since January 2026, with ongoing updates throughout the year. The feature utilizes three AI models to convert faces into biometric data and is integrated into the Meta AI companion app, which is essential for the glasses' functionality. Although the feature is not yet operational, Meta has committed to transparency regarding its potential rollout and has stated it will not establish a central face database.

Anthropic Calls for AI Pause Button to Let Humans Take Stock

Anthropic PBC has proposed an industrywide mechanism to pause artificial intelligence development, allowing society to evaluate its profound implications. In a blog post by co-founders Jack Clark and Marina Favaro, the company highlighted that AI advancements could lead to human labor becoming thousands of times more efficient or even entirely replaced. This call for a pause, made on June 5, 2026, underscores the urgency of addressing the associated risks before further progress is made.

Rocket Report: Blue Origin explosion still making headlines; Impulse raises money

Blue Origin's New Glenn rocket exploded on May 28 at its launch pad in Florida, raising questions about the company's timeline for resuming operations. CEO Dave Limp stated that they aim to launch again by the end of 2026, despite skepticism from industry veterans regarding the feasibility of this timeline. Meanwhile, Impulse Space announced it raised $500 million in Series D funding, bringing its total funding to over $1 billion since its inception. The Canadian government is also investing $182.6 million CAD to develop a sovereign launch capability, highlighting a growing interest in space infrastructure.

How C3 AI agents will automate predictive maintenance for Shell

Shell is set to enhance its predictive maintenance capabilities by deploying C3 AI agents, transitioning from basic anomaly detection to a fully automated maintenance lifecycle. This initiative builds on the existing C3 AI Reliability Suite, which currently monitors over 30,000 critical assets across Shell's operations. The new system will allow AI agents to autonomously investigate alerts, draft work orders, and manage procurement, thereby minimizing the need for human intervention. This shift aims to significantly reduce unplanned downtime and improve operational efficiency, ultimately delivering hundreds of millions of dollars in economic value.

The Meta hack shows there’s more to AI security than Mythos

On June 5, 404 Media reported that attackers exploited Meta's AI customer support agent to hijack Instagram accounts by simply requesting the agent to link them to email addresses they controlled. This method led to the takeover of several accounts, including a dormant Obama White House account, which was then used to post pro-Iran content. Experts like Neil Gong from Duke University expressed surprise that such a straightforward exploit went unnoticed by Meta, particularly given the company's extensive expertise in AI and cybersecurity. The incident highlights significant vulnerabilities in AI systems, raising concerns about the security of automated workflows as companies increasingly rely on AI.

Google Chrome tests sending users straight to AI Mode instead of Search

Google is testing a new feature in Chrome that directs users straight to AI Mode, bypassing the traditional Search experience. This functionality, identified in the latest Chrome Canary build, allows users to enter AI Mode by typing any query in the omnibox, a significant shift from the standard Google Search interface. However, this feature is not enabled by default and requires users to manually activate it through chrome://flags. A Google commit clarifies that this is merely an exploratory phase, with no immediate plans to make this change permanent, suggesting that the traditional Search experience remains intact for now.

Presentation: Platform Teams Enabling AI - MCP/Multi-Agentic Tools Across Linkedin

LinkedIn is implementing AI as a new execution model to enhance engineering efficiency, driven by its vast scale of operations, which includes 1.3 billion members and 45 trillion Kafka messages exchanged daily. Karthik Ramgopal and Prince Valluri outlined how AI can replace fragmented implementations with structured abstractions for orchestration and tooling. By focusing on areas where tasks are repeatable and outcomes easily validated, LinkedIn aims to automate coding, migrations, and testing, thereby reducing the cognitive load on its 10,000 engineers and improving overall productivity.

Dropbox Introduces Nova, an Internal Platform for Running AI Coding Agents at Scale

Dropbox has launched Nova, an internal platform designed to orchestrate AI coding agents across its engineering workflows. Unlike standalone coding tools, Nova provides a centralized execution layer that integrates with Dropbox's monorepo, CI systems, and operational tooling. This allows AI agents to engage in tasks such as flaky test remediation and dependency migrations, operating safely within a customized environment. Each Nova session is isolated and tied to specific repository commits, enabling a 'propose, validate, iterate' workflow that enhances reliability and efficiency in software engineering.

OpenAI and Anthropic May Be Rivals, but Investors Aren’t Picking Sides

OpenAI and Anthropic, despite being fierce rivals in the AI sector, share a significant overlap in their investor base, with about 90 venture capital firms backing both companies. According to PitchBook data, approximately 42% of OpenAI's investors are also investors in Anthropic, including major firms like Sequoia Capital and Greylock. Anthropic recently announced a fundraising round with 31 investors, of which at least 13 have stakes in OpenAI. This unusual investment strategy reflects a changing landscape in venture capital, where firms are increasingly willing to invest in competing companies as both OpenAI and Anthropic prepare for potential IPOs.