[AI DAILY NEWS RUNDOWN] The Death of Search Traffic, Anthropic’s Agent Limits, and Cerebras’ $107B IPO (May 14 2026)

The Rundown: Fintech firm Ramp just published its latest AI Index, showing Anthropic taking the paid business-adoption lead from OpenAI for the first time, an enterprise surge that has quadrupled its usage over the past year while OpenAI has leveled off.

The details:

• Ramp tracks corporate card and invoice payments from 50K+ U.S. businesses, making this a spend signal, not full market share.

• Anthropic rose 3.8% in April to 34.4% of adoption, while OpenAI fell 2.9% to 32.3% as overall AI use continued to climb to 50.6%.

• Claude Code anchored much of the swing, with Anthropic expanding from technical teams into finance, legal, and research workflows.

• Ramp highlighted several risks facing Anthropic despite the trend, including recent Claude outages and increasing costs compared to OAI and open-source.

Why it matters: OpenAI is not suddenly cooked, with Ramp not tracking some large enterprise deals and ChatGPT remaining the bigger consumer brand. The yearly trend speaks louder, and was likely the same type of chart that caused OAI’s Fidji Simo-led pivot, which has become evident from recent Codex and other enterprise pushes.

Amazon doubles down on Alexa+ for shopping

Image source: Amazon

The Rundown: Amazon folded its standalone shopping chatbot ‘Rufus’ inside “Alexa for Shopping”, a new agent that takes over Amazon search and follows shoppers across devices with a shared memory of purchases, preferences, and prior chats.

The details:

• Rufus drew 300M+ users in 2025 while still in beta, with its product knowledge and shopping history now feeding Alexa for Shopping answers.

  
  
  

• Amazon says the new assistant draws on catalog data, reviews, delivery timing, past purchases, and Alexa conversations for information.

• Alexa can now field questions in the search bar, run side-by-side comparisons, track pricing, and Auto-Buy items when prices hit a target.

• A new Buy for Me handles checkouts on non-Amazon stores, with Scheduled Actions able to automatically restock products on a cadence.

Why it matters: RIP to Rufus, but consolidating under one Alexa AI brand feels like a better play, and the sheer amount of customer history Amazon has gives its agent quite the moat to work with. But with a consumer base that is already moving to other AI platforms for their agentic shopping needs, will Alexa play nice and integrate with them?

Leo sets Catholics on collision course with AIBy Russell Contreras

Pope Leo XIV is expected to sign his first encyclical as soon as tomorrow, positioning AI as the defining moral and labor challenge of a new industrial revolution.

Why it matters: The document, reportedly titled “Magnifica Humanitas” (”magnificent humanity”), would become the Catholic Church’s clearest attempt yet to place human dignity, labor rights and ethics at the center of the AI race.

• Catholic and European outlets are reporting that Leo is poised to sign the AI encyclical on the anniversary of “Rerum Novarum” (1891), Pope Leo XIII’s foundational industrial-era labor encyclical.

• The encyclical will focus specifically on AI’s impact on “people and working conditions,” framing it as Leo XIV’s effort to modernize Catholic social teaching for the AI era, per the French newspaper Le Monde.

• Other reports suggest “Magnifica Humanitas” will argue technology must remain subordinate to the human person — not the reverse — and that AI systems should protect workers, creativity and moral agency.

The Vatican has not commented, but it has implemented formal AI guidelines and monitoring structures inside Vatican City.

Exclusive: 6-step technique to get Mythos-proofed

With AI coding tools poised to transform the job of a software engineer, the question remains: how trustworthy are these models’ outputs?

While vibe coding is helping engineers be more productive than ever, the code itself isn’t always as secure as it could be. And as models become more powerful, those vulnerabilities can be easily exploited: Data from Experian, published last week, found that 40% of the 5,000 data breaches it handled last year were driven by AI. With AI creating a cybersecurity powder keg, Synthesia decided to do something about it.

On Thursday, the company published a blog post detailing its approach to automating code security reviews, offering a recipe that any company can use and retrofit to generally available models. The company claims this six-step technique achieves cybersecurity performance similar to Anthropic’s Mythos.

To break it down simply:

• The first two steps involve prepping the code to identify potential vulnerabilities and mapping every possible entry point for untrusted input.

• For the next two steps, the system sends AI agents to hunt for three types of vulnerabilities: injection attacks, authorization issues, and business logic flaws. Duplicate findings are removed, saving tokens by avoiding the need to thoroughly investigate them.

• Finally, the AI agents first double-check each vulnerability and then summarize what actually needs to be fixed.

Martin Tschammer, head of security at Synthesia, told The Deep View that this technique is a solution to a problem that the company aimed to tackle internally first. As a software firm, it’s using AI and coding agents to ship products faster than ever.

“Security review in particular depends on careful, deliberate analysis so it’s exactly the kind of work that breaks down under velocity pressure,” Tschammer said.

Tschammer also said this system offers several particular advantages. For instance, rather than just understanding general code, applying this to the models you use in your enterprise means they understand your specific codebase.

It also cuts out additional work before reaching your team, and doesn’t require a frontier model to leverage. Plus, at an average of less than $4 per review, it’s an affordable way to prevent vulnerabilities from slipping through the cracks.

“You do not have to wait for Mythos, or some other security-branded model, to become available,” Tschammer said. “In fact, we believe you don’t have the luxury to wait. If you want to stay ahead of the curve, you need to act now.”

Adaption automates AI training with AutoScientist

Image source: Adaption

The Rundown: Adaption, the AI startup from ex-Cohere VP of Research Sara Hooker, just introduced AutoScientist, a new system that automatically customizes AI models for specific jobs by adjusting both what the model learns from and how it learns.

The details:

• AutoScientist tests different training data and settings, then iterates until the model meets the user’s goal.

• In internal tests, AutoScientist outperformed its own expert-tuned models by 35% on average, with success rates jumping from 48% to 64%.

• Results held across multiple AI models, a wide range of dataset sizes, and 8 varied industries, including finance, legal, and medical domains.

• Adaptive’s initial Adaptive Data release in February aimed at increasing the quality of datasets, with Autoscientist now moving toward customizing models.

Why it matters: A few thousand people in the world know how to properly train and fine-tune a frontier model, and nearly all of them work at the same handful of labs. If a tool like Autoscientist can start to automate that expertise, models customized for individual businesses and use cases may become a lot more practical to create.

US approves 10 Chinese firms to buy Nvidia H200 chip

• The US has approved sales of Nvidia’s H200 AI chips to 10 Chinese companies, though not one chip has actually shipped since President Trump first cleared the deal back in January of this year.

• Nvidia CEO Jensen Huang was added at the last minute to Trump’s Beijing trip, which a former Taiwan legislator said shows the president wants Huang at the table to serve as a bargaining chip in talks.

• Chinese tech firms are leaning more on homegrown semiconductors, signaling to Washington that export curbs haven’t slowed China’s AI progress, while American voters souring on AI could make White House backing for Nvidia harder to defend.

Instagram takes on Snapchat with disappearing photos

• Instagram is rolling out “Instants” worldwide, a new feature that lets users share disappearing photos with Close Friends or mutual followers, putting it in direct competition with Snapchat after a testing period with select users.

• Photos must be taken with Instagram’s in-app camera with no edits or uploads allowed, can be viewed only once, stay available for 24 hours, and recipients can’t screenshot or record them.

• Shared Instants sit in a private archive for up to one year, can be compiled into a recap for Stories, and senders can tap “undo” to unsend an Instant before friends open it.

LinkedIn cuts 5% of its workforce

• LinkedIn is laying off 5% of its staff as the company shifts its workforce toward parts of the business that are growing, according to people who spoke with Reuters about the restructuring plans.

• The cuts will affect a sizable share of LinkedIn’s more than 17,500 full-time workers globally, with the move tied to a broader push across the platform to reshape operations around AI.

• LinkedIn’s announcement adds to a heavy year for tech layoffs, with Layoffs.fyi tracking over 103,000 job cuts so far in 2026, closing in on the roughly 124,000 cuts recorded throughout all of 2025.

Cerebras IPO doubles valuation basically overnight

Chip design company is Cerebras trading at $350 per share on its first day of public trading, which values the company at ~$107B and has them raising $10.5B. The price on this IPO has literally been up only. On Monday, price range for the shares were updated to $150-$160 from $115-$125, meaning the company’s market cap has nearly doubled overnight.

SemiAnalysis has a fantastic deep dive on Cerebras that’s worth reading in full.

The interesting takeaways are some really solid positives. Cerebras chips work, they are very fast, and token-consumers have revealed a preference and willingness to pay for speed. The SemiAnalysis team quantified it in their own experience with Opus models:

Opus 4.6 fast mode famously charges 6x the price for 2.5x the interactivity (though it’s now under 2x faster, see chart below). In April, 80% of our AI spend (which peaked at $10M annualized) was on Opus 4.6 fast.

Cerebras has a big 750mw deal with OpenAI and chips are already serving GPT-5.3 in Codex on Cerebras chips via “Spark” – it’s remarkably fast, and even if you aren’t vibe coding, you can just go talk to the model like it’s ChatGPT and experience the difference. I think there will be huge demand for faster inference across all parts of the AI economy. There’s an old (probably slightly apocryphal) quote about ecommerce that goes: “Every 100ms of latency costs Amazon 1% in sales.” This is currently playing out in AI inference. Companies pay disproportionately more for faster inference. Good for Cerebras.

SemiAnalysis does point out a bunch of problems the team will have to solve or contend with over the next few years though. Mainly, Cerebras chips are not currently as capable of holding larger models or networking multiple chips together to serve large models and handle larger context windows:

“Moreover, the industry is trending towards larger context windows ad infinitum — 128k context will certainly not be acceptable for long, especially with the prevalence of agentic workloads.”

And it doesn’t look like there is a simple solution of just scaling the wafer size larger or adding more memory to the existing architecture. Cerebras’s whole design depends on lots of SRAM directly on the wafer. But SRAM is no longer shrinking much with each new semiconductor node. SemiAnalysis points out that WSE-2 had 40GB, WSE-3 only moved to 44GB, and future nodes may not help much. That means Cerebras cannot simply wait for Moore’s Law to give it much more memory. To add more SRAM, it may have to sacrifice compute area or attempt much harder 3D/wafer-bonding approaches.

In an agentic workflow, it’s entirely possible you will want the biggest, most powerful model handling critical work, so future models might not fit neatly on Cerebras chips. Still, I imagine that we will quickly jump from the agentic age to the orchestrator age and there will be hybrid approaches. The biggest and best models will delegate certain tasks to smaller, faster models, while they crank on harder problems. It’s hard to predict the exact mix of chips that will power large networks of agents, but these different designs seem more complementary than directly competitive. — John

General Catalyst ad raises eyebrows on the timeline

​​GC took direct shots at Marc Andreessen yesterday in a new ad called Meet GC. If you weren’t aware of Marc’s love of robot dogs, you might not have put it together. But Marc quickly responded dozens of times, at one point calling it “a clever inversion of the original ‘Mac vs PC’ commercials, this time clearly intended to make the sponsor appear smarmy and judgmental. Fascinating.”

It’s fascinating to me to see GC try to position themselves on the moral high ground over a16z, considering that GC has backed both Kalshi and Polymarket, two of the companies at the very center of tech’s morality debate today.

The winner here is no one. I think both sides would have more aura if they focused on being more like Blackstone. Run ads in The Economist, WSJ, and Financial Times, founders do an interview once a year, and otherwise focus on themselves. — Jordi

Anthropic tightens Claude limitsBy Ina Fried

Anthropic is putting new limits on what paying customers can do with their subscriptions, giving rival OpenAI an opening to lure power users to Codex.

Why it matters: The fight shows that “all-you-can-eat” AI subscriptions may not survive the agent era, where software can burn through computing resources far faster than humans ever could.

Driving the news: Anthropic announced that it’s bringing back support for outside agent tools on paid Claude plans. But it is putting that usage behind a separate credit meter.

• Subscribers will now get a new monthly credit that they can use with third-party harnesses like OpenClaw.

• Anthropic says that new changes should support the ways that the majority of people use Claude.

What they’re saying: Anthropic’s changes didn’t go over well.

• Claude Code product manager Noah Zweben’s X post about the new rules was riddled with critical replies, with respondents calling the changes “gaslighting” and claiming to be switching to Codex.

The intrigue: OpenAI is taking the opposite tack, at least for now. CEO Sam Altman announced on X that OpenAI is giving new business customers two months of free Codex usage.

Zoom in: The industry appears to be rediscovering a lesson from earlier eras of computing: humans have built-in limits to how much data they can consume, while automated workloads can explode usage.

• A human might send dozens, or perhaps hundreds, of prompts a day, while an autonomous coding agent can generate thousands of requests, run tests continuously, browse the web, and recursively call models.

The other side: Businesses are finding that AI agents can lead to hefty bills.

• ServiceNow and Uber are among the companies that have already burned through their AI token budgets for the entire year, per The Information’s Laura Bratton.

What we’re watching: Everyone is facing the same economics and will eventually need to move away from unlimited use.

• Anthropic has been among the more aggressive in restricting use because it’s been the top choice for coders who use agents the most and have been struggling to maintain enough compute resources.

Kevin Warsh is confirmed as the new Fed Chair right after inflation rips

Trump’s pick to replace Jerome Powell was confirmed as the next Fed Chair yesterday in a 54-45 Senate vote. Current chair Jerome Powell ends Friday. The expectations for Warsh have been that he’ll be more open to cutting interest rates, and his background — he’s pro Bitcoin and AI and has VC exposure to crypto — aligns with the current Administration. But he’s also stepping into the role just as we had our biggest inflation print since May 2023, which might immediately put him at odds with the President on the rate cut issue.

Elon-OpenAI trial expected to end today

Closing statements started this morning in Oakland in Elon Musk’s court case against OpenAI, in which he’s accusing the company of ‘stealing’ a charity for for-profit conversion. The jury will have to deliberate on the case, so it would be surprising if the judge handed down a decision today. Also worth remembering that the jury in this case is advisory, so Judge Yvonne Gonzalez Rogers retains authority over remedies and certain final decisions herself.

Clip Spotlight: Condé Nast CEO Says Search and Social-Based Digital Publishing Is Over

After several years of declining search traffic, Condé Nast CEO Roger Lynch has directed all the company’s brands to operate as if search traffic to their properties will be zero.

He says the era of turning search and social media traffic into profitable businesses is gone.

And that if you run a media business that doesn’t have an authoritative brand, a very strong niche, or a direct audience, you’re going to be fighting hostile algo changes all the way down.

He describes a recent board meeting:

“We took a snapshot of search results from seven or eight years ago. And what you saw were a few sponsored links, then the ten blue links.”

“Do the same search today, you get an AI overview, then you get rows and rows and rows of commerce links, then you get sponsored stuff.”

“Each of the last three years, we would do our budgets, and we’d put forecasts in of search traffic declining. Because we’d seen the pattern of algorithm changes. And generally those algorithm changes were negative.”

“Every year, our search traffic was down more than we had forecast. So last year I told our teams, ‘Assume there’s no search.’ You have to have your businesses planned as if search is zero. We don’t expect it to be zero, we expect it to be a single-digit percentage of our traffic.”

What Else Happened in AI on May 14th 2026?

Nvidia became the first company to hit a $5.5T market cap, coming as CEO Jensen Huang arrived in China to join U.S. President Donald Trump in meetings with Xi Jinping.

Sam Altman testified in the Elon Musk vs. OpenAI legal battle that Musk’s “specific plans on safety” made him worry, and proposed passing the company to his children.

David Silver’s Ineffable Intelligence and Nvidia announced a partnership to build training pipelines for RL agents, with early work targeting Nvidia’s Vera Rubin hardware.

Microsoft introduced MDASH, an AI security harness that chains 100+ specialized agents to hunt for software bugs, with the system catching 16 flaws across Windows.

UK’s AI Safety Institute said that AI’s ability to complete cyberattacks is doubling every few months, with Mythos Preview and GPT-5.5 finishing its simulated breaches.