☕️ OpenAI weighs price cuts to beat Anthropic

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

💸 OpenAI weighs price cuts to beat Anthropic

🔧 Anthropic backtracks on policy that 'sabotaged' researchers' work

🎮 Xbox warns of a ‘reset’ as it prepares for layoffs

❌ Judge denies Meta and Google new trial

💳 Visa plugs its payment network into ChatGPT

Plus: 🎁 12 other news you might like, 🧰 6 tools, and 📚 5 papers.

💸 OpenAI weighs price cuts to beat Anthropic LINK

OpenAI is considering big cuts to what it charges for AI tokens, partly because it expects rival Anthropic to lower its own prices first and wants to be ready to respond.

CEO Sam Altman has called AI costs "a huge issue" for business customers, but cutting token prices would squeeze margins as both companies already lose billions and pursue IPOs that expose their finances to public investors.

The pressure follows Anthropic's revenue surge from its coding tool Claude Code, which briefly pushed its valuation past OpenAI's and prompted OpenAI to make its own coding tool, Codex, a priority.

🔧 Anthropic backtracks on policy that 'sabotaged' researchers' work LINK

Anthropic is reversing a hidden policy that quietly limited researchers who used its new Claude Fable 5 model to build competing AI systems, telling Wired the company "made the wrong tradeoff" and apologizing for the secrecy.

Researchers found that Fable 5, built on Anthropic's Mythos system, would quietly reroute requests to a weaker model or degrade responses for tasks like training competing LLMs, debugging AI code, and optimizing neural architecture.

Anthropic isn't dropping the safeguards but making them visible, so if it suspects a user is building a highly capable AI, it will alert them that it's refusing the request or rerouting to a less capable model.

🎮 Xbox warns of a ‘reset’ as it prepares for layoffs LINK

Xbox bosses Asha Sharma and Matt Booty have warned staff in a memo of an "Xbox reset" over the next 100 days, coming ahead of major layoffs expected in July.

The memo said that, excluding Activision Blizzard King, Xbox has spent over $20 billion on content, platform, and hardware subsidy in five years, while annual revenue dropped nearly half a billion.

Sharma and Booty also pointed to a hardware component crisis, saying costs for the 2027 holiday season are expected to top five times the prices paid two years earlier, prompting calls for a new business model and hardware partnerships.

❌ Judge denies Meta and Google new trial LINK

A California state court judge has rejected requests by Meta and Google's YouTube for a new trial, after a jury found both companies liable for designing social media platforms that harm young people.

The jury decided Meta and YouTube were negligent in operating their platforms, harming a 20-year-old plaintiff named Kaley, and assigned Meta 70 percent of the responsibility compared to 30 percent for YouTube.

Judge Carolyn Kuhl dismissed the argument that Section 230 shields the companies, writing the law does not cover their design choices, and both Meta and Google said they plan to appeal.

💳 Visa plugs its payment network into ChatGPT LINK

Visa is working with OpenAI to plug its payments tools into OpenAI's agent system, letting AI agents spend money for users and complete transactions without people manually checking out every time.

Users can set spending caps, merchant restrictions and approval requirements, while Visa handles fraud detection, chargebacks and refunds, and the companies say the setup could support shopping, business invoice payments and AI coding agents buying APIs.

Visa's Rubail Birwadker said more than one in five transactions are influenced by what users learn through LLMs, and the experience could end up feeling similar to shopping with Apple Pay or Shop Pay.

Other news you might like

• OpenAI says China targeted America’s AI future with covert influence campaignLINK

• Anthropic’s Dario Amodei wants governments to have the power to block ‘dangerous’ AI systemsLINK

• Deezer launches an AI music detector for other streaming servicesLINK

• Humanoid robotics company raises up to $1.4 billion from Nvidia, Amazon and othersLINK

• Oracle beats on earnings, but stock drops on plans to raise another $20 billionLINK

• You can just tell the Instagram algorithm what you want nowLINK

• The Trump Mobile T1 really is an HTC phone from 2024, teardown showsLINK

• Meta has cut Manus off from its internal systems and told staff to sunset the AI platformLINK

• Prometheus, Jeff Bezos' AI startup, is now worth $41 billionLINK

• ‘AI-pilled’ firms spend $7,500 per employee each month on AILINK

• Anthropic’s Dario Amodei has just one direct reportLINK

🧰 Trending tools

Terminal Mode by Even Realities: lets developers monitor AI coding agents, approve actions, and stay in control of automated sessions directly through smart glasses.LINK

Journey Now: a learning planner that creates adaptive 30-day roadmaps with daily exercises, weekly recaps, and friend tracking to maintain consistent progress toward any goal.LINK

CrustRecruiter: real-time API that tracks job changes, promotions, new skills, and company activity so recruiters never miss a hiring signal.LINK

Tabstack Structured Extraction: converts any webpage into structured JSON by matching your custom schema, removing the need to write or maintain scrapers.LINK

Cloudskill: centrally manages AI agent skill files with version control, approval workflows, access policies, and audit logs to reduce security risks and workflow chaos.LINK

Lium Ai: lets domain experts query large multimodal datasets using plain English, converting complex data analysis into shareable, reusable team workflows.LINK

📚 Trending papers & reports

"AI slop" accusations have exploded tenfold online but do not actually identify AI writing, instead functioning as social gatekeeping against anything that feels inauthentic, regardless of who wrote it.LINK

Startup delays in AI serving software have been mapped for the first time, revealing what causes them and letting engineers accurately predict how long a system will take to become ready.LINK

Writing and thinking tools built on large language models may quietly push everyone toward the same words and ideas, flattening the human diversity of thought that fuels creativity and problem-solving.LINK

The goal of "general AI" is the wrong target, this paper argues, proposing instead that AI should aim to be superhuman at specific tasks and fill gaps where humans simply cannot perform.LINK

Biology lab robots can now be programmed by AI agents that outperformed the median expert human on every tested task, including assembling real DNA correctly in physical lab experiments.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!