How AI could be causing higher gas prices

PRODUCTS

OpenAI just put its AI to work fixing the bugs that hold the internet together.

On Monday, OpenAI launched Patch the Planet, a new initiative that uses its models to find security flaws in open-source software and actually write the fix. The company is running the effort with Trail of Bits, a security research firm, and the early numbers are doing the talking.

In its first week across 19 open-source projects, the results already stand out:

+ 64 pull requests filed, 51 issues raised, and 19 already closed with a fix

+ The new GPT-5.5-Cyber model scored 85.6% on CyberGym, a benchmark for real-world security tasks — edging out Anthropic's Claude Mythos Preview at 83.8%

+ Roughly half the AI's time went to writing fixes, not just finding bugs — which is the part open-source projects have never had enough hands for

That last point is what makes this different from earlier AI security tools, which mostly surfaced vulnerabilities and dumped them on overworked maintainers to patch. Patch the Planet is built to do the whole job: find the bug, validate it, write the fix, and explain the change.

The timing matters, too. Anthropic has been running a parallel effort called Project Glasswing with bigger raw numbers — more than 10,000 flagged vulnerabilities across roughly 50 partners. But Anthropic recently had to suspend access to its Mythos model after a US government export-control directive, putting parts of Glasswing on pause. OpenAI is taking ground while its main rival is grounded.

The catch is that open-source code runs nearly everything — including the systems AI agents use to write new software — and the people maintaining it are wildly outnumbered by attackers looking for ways to break it. Flooding those same maintainers with AI-generated patches could just move the bottleneck from "not enough fixes" to "not enough reviewers." A few open-source maintainers have already asked Anthropic to slow down disclosures because they need more time to evaluate patches.

The story most people will tell is that OpenAI and Anthropic are racing to build the better security AI. The more interesting one is that they've both quietly become responsible for the security of open-source code they didn't write. That's a strange spot for a company whose main product is a chatbot, and it's going to get stranger as these systems file more patches in a week than maintainers can review in a month. The next real question isn't whether the AI can find the bugs. It's whether anyone on the other end has time to read what it sends.

ENTERPRISE

Three years ago, Samsung banned ChatGPT after engineers pasted confidential source code into the public version, risking their proprietary IP being absorbed into OpenAI's training data. The company blocked it company-wide and tried building internal AI tools instead. None gained real traction.

Now Samsung is going all-in on the thing it once feared. On Monday, OpenAI announced that Samsung is rolling out ChatGPT Enterprise and Codex to employees worldwide — from chip fab engineers to marketing teams — calling it one of the largest enterprise deployments in OpenAI's history.

The reversal makes sense because ChatGPT Enterprise solves the exact problem that triggered the ban: customer data isn't used to train OpenAI's models, and companies get full admin controls over what employees can do with it.

But the more interesting part of this deal isn't the chatbot — it's Codex. Samsung recently announced a plan to transition its global manufacturing into "AI-driven factories" by 2030, and Codex is positioned as the layer that makes that possible. Codex weekly active users in Korea have grown nearly 800% since February, with more than 5 million users globally. Samsung's vision is that Codex moves beyond helping developers write code to eventually making autonomous decisions on the factory floor — AI that "understands operational contexts in real time and independently executes optimal decisions," as one Samsung EVP put it.

For OpenAI, this is the payoff for the enterprise sales machine it's been quietly building — poaching from Salesforce and acquiring Ona earlier this year to make Codex enterprise-ready with the security controls companies like Samsung require.

The most interesting line in this whole deal is the one about autonomous factories. Selling ChatGPT to office workers is the boring part of the enterprise AI story, and most of the big rollouts so far have looked roughly the same: emails get written faster, slide decks come together quicker, nothing fundamental changes. Samsung is pointing at something different. If Codex eventually moves from helping engineers debug code to making decisions on a chip fab floor, that is the version of enterprise AI that actually rewrites how a company operates. Whether OpenAI can deliver on that is a much bigger question than whether Samsung's marketers like the new chatbot.

GOVERNANCE

A new class action accuses BP, Marathon, 7-Eleven, and Walmart of using the same AI pricing software to keep California gas prices artificially high — and the legal theory behind it should sound familiar.

The lawsuit, filed in California state court last week, claims all four companies licensed pricing tools from a firm called Kalibrate, which ingests competitor data and recommends daily price points. The plaintiffs' argument is straightforward: when rivals feed their numbers into the same algorithm and follow its suggestions, that's not competition — it's a quieter version of price-fixing.

If that sounds like the RealPage rental case, it should. The DOJ spent two years going after RealPage's apartment pricing software on nearly identical grounds, and it worked:

+ RealPage recently agreed to stop sharing competitor data after DOJ action. Private landlord settlements have totaled around $360 million.

+ The algorithm's influence was massive — nearly 60% of final apartment prices landed within 2.5% of what the software suggested.

+ The DOJ has explicitly warned this would spread, calling algorithmic price-fixing "old crime, new code" and pledging to stay at the forefront of enforcement as AI pricing tools proliferate.

The defense will be what it always is: Kalibrate's tools only offer recommendations, and companies are free to ignore them. But California is a tough venue to make that argument. The state AG's office just settled a separate algorithmic rent-pricing case for $7 million, and California's gas market is already among the most scrutinized in the country thanks to a transparency law passed in 2023.

The interesting thing about this lawsuit is not whether the gas station chains lose. It is that the legal theory behind it is now portable. Once a court accepts that rivals quietly feeding their numbers into the same algorithm can look like a conspiracy, the playbook works on hotels, airlines, insurance, anywhere a single software vendor has become the de facto pricing brain of an industry. Companies that have spent the last few years quietly outsourcing their pricing decisions to AI vendors are about to find out whether they bought a productivity tool or a legal exposure. The defendants will argue the algorithm is just a suggestion. Regulators have already decided that a smoke-filled room with a Wi-Fi connection is still a smoke-filled room.

IN OTHER NEWS

+ Oracle cuts 21,000 jobs — 13% of its workforce — and says AI layoffs will continue

+ China reclaims the world's fastest supercomputer crown with a machine built entirely without Nvidia GPUs

+ SpaceX signs a $6.3 billion compute deal with Reflection AI , a two-year-old open-source startup backed by Nvidia

+ A $2B robot startup is secretly renting Airbnbs to test household robots — and trashing them in the process

+ Menlo Ventures raises a $3 billion fund after its Anthropic bet turned $750M into $14B

+ Leaked memo reveals Meta's massive AI reorg backfired — CTO Bosworth admits morale is "the worst it's ever been"

+ Music platform Jamendo sues Nvidia for training AI models on 55,000 songs without permission — Nvidia's 8th copyright suit

+ Trump signs two executive orders to fast-track quantum computing , targeting a practical breakthrough by 2028

WHO'S HIRING IN AI

+ Anthropic — Technical Program Manager, Launches

+ Salesforce — Director of Product Management, Agentic Workflows

+ Google — Product Manager, Advanced Enterprise Data Platform

+ Airwallex — Product Lead, Agentic Finance

AI OR REAL?

Which image is real?

Option A | Option B

AI TOOLS

+ Claude Tag: Anthropic's new always-on Slack teammate follows your team's conversations, builds context over time, and jumps in to handle tasks when you @mention it — available now for Enterprise and Team customers

+ Adobe Firefly: Adobe's AI assistant now lives inside Photoshop, Premiere, Illustrator, and InDesign — tell it what you want in plain English and it handles multi-step edits across your creative workflow

+ Gemini on Android: Google expanded the Gemini overlay with video generation, music creation, and guided learning tools you can access from anywhere on your phone without opening the full app

+ Runway Aleph 2.0: Runway's AI video editing model is now inside Figma Weave — edit a single frame and the changes automatically carry through your entire video

+ Krea 2: A new open-weight image model that generates enterprise-grade visuals in 2 seconds — designed to look less like "AI slop" and more like actual creative work