ChatGPT reaches for your bank account; OpenAI reorgs again

OpenAI launched a personal-finance experience that connects directly to bank accounts, the same week the CFTC said it will use AI to police insider trading on prediction markets and Databricks made GPT-5.5 its default enterprise agent model. Greg Brockman took unified control of OpenAI products in another executive reshuffle, and the company is reportedly preparing legal action against Apple over the failed ChatGPT integration. The information-integrity story hardened: arXiv will ban researchers who submit AI-generated slop, a widely-shared post argued frontier AI has broken the open CTF scene, and a new paper proposes a benchmark for predicting scam progression before victims commit. The data-center backlash spread from Lake Tahoe to a Pennsylvania town hall — opposition is now a recurring local-news beat, not an outlier. Mitchell Hashimoto's claim that 'entire companies' are operating under AI psychosis pulled 1,430 points on Hacker News; an Andon Labs experiment running AI radio hosts gave the same warning in production.

8 papers 22 news 9 sources ← Latest

News

18 items

AI reaches for your money

OpenAI shipped a personal finance experience in ChatGPT with direct bank-account connection, Databricks made GPT-5.5 its default for enterprise agents, and the CFTC said it will use AI to police prediction markets. The throughline: AI is moving from advising on money to handling it.

News TechCrunch AI

OpenAI launches ChatGPT for personal finance, will let you connect bank accounts

ChatGPT Pro users in the US can now connect bank and brokerage accounts and see a unified dashboard of portfolio, spending, and subscriptions.

Why it matters

Moves ChatGPT from chat surface to financial UI — a direct shot at Mint, Copilot Money, and bank apps.
Establishes Plaid-style banking connectivity as table stakes for consumer AI assistants.
Sets up a clear consent and liability stack the rest of the industry will have to match.

Source →

News The Verge AI

OpenAI now wants ChatGPT to access your bank accounts

The Verge's read on the same OpenAI launch — frames the bank-connection as the moment trust in AI is operationally tested.

products safety regulation

Source →

News OpenAI

A new personal finance experience in ChatGPT

OpenAI's official launch post detailing the Pro-tier personal finance feature and the security framing around financial-account connection.

products

Source →

News OpenAI

Databricks brings GPT-5.5 to enterprise agent workflows

Databricks adopts GPT-5.5 for enterprise agent workflows after it set state-of-the-art on the OfficeBench-style benchmark.

Why it matters

Anchors GPT-5.5 in the data-warehouse layer where most regulated enterprise AI work lives.
Direct counter to Anthropic's recent business-customer lead — a credible enterprise-agent default.

products agents market

Source →

News Ars Technica AI

The US is betting on AI to catch insider trading in prediction markets

The CFTC announced it will use AI to detect insider trading on prediction-market platforms — surveillance AI lands inside a federal regulator.

regulation policy safety

Source →

The information-integrity crisis tightens

arXiv said it will ban researchers who upload AI-generated slop, a viral essay argued frontier AI has broken the open CTF format, and a new paper proposes predicting scam progression before victims commit. Across academia, security, and consumer scams, the gatekeeping infrastructure is having to be rebuilt for an AI-saturated information environment.

News The Verge AI

ArXiv will ban researchers who upload papers full of AI slop

arXiv is moving to ban submitters who upload AI-generated hallucinated content — a yearlong vacation from submissions for offenders.

Why it matters

First credible enforcement action by a major preprint platform against AI-generated noise.
Sets a precedent other archives and journals will be pressured to follow.
Concrete evidence the volume problem is now severe enough to warrant exclusion.

policy regulation data

Source →

News Ars Technica AI

Send the arXiv AI-generated slop, get a yearlong vacation from submissions

Ars Technica details arXiv's new submitter-ban policy as announced by a site moderator.

policy data

Source →

News Hacker News

Frontier AI has broken the open CTF format

Widely-shared argument (160 HN points) that frontier AI has made traditional open capture-the-flag competitions unwinnable for human teams.

safety evaluation community

Source →

News The Verge AI

YouTube is expanding its AI deepfake detection tool to all adult users

YouTube opens its AI likeness-detection program to all users over 18, the largest platform-level deepfake control rollout to date.

safety products

Source →

OpenAI's structural week

Greg Brockman took unified control of OpenAI products in another executive reshuffle, the Musk v. Altman jury began deliberating, and OpenAI is reportedly preparing legal action against Apple over the failed ChatGPT integration. The company is restructuring while litigating its past and arming for its next platform fight.

News Wired AI

Greg Brockman Officially Takes Control of OpenAI's Products in Latest Shake-Up

OpenAI consolidates ChatGPT and Codex under Greg Brockman's product leadership in another internal reorganization.

Why it matters

Unifies the two surfaces OpenAI uses to reach paying users and developers.
Signals the agent fight is now treated as a product, not a research, problem.

products market

Source →

News The Verge AI

OpenAI keeps shuffling its executives in bid to win AI agent battle

The Verge's read on OpenAI's reorg as a response to the agent-battle competitive pressure from Anthropic and Google.

market agents

Source →

News Ars Technica AI

OpenAI feels 'burned' by Apple's crappy ChatGPT integration

Insiders say OpenAI feels burned by Apple's ChatGPT integration; separately, a judge ordered Apple to hand internal messages over to Musk.

market policy regulation

Source →

News MIT Technology Review

Musk v. Altman week 3: Now the jury will pick

Final-week recap: lawyers traded credibility blows and the case is now with the jury.

policy market

Source →

News Ars Technica AI

Anthropic's $1.5B copyright settlement is getting messy as judge delays approval

Authors are fighting for higher payouts and accusing class lawyers of rushing the $1.5B Anthropic copyright settlement to capture $320M in fees.

settlement size $1.5Bdisputed fees $320M

regulation policy market

Source →

Data center backlash spreads

Pennsylvanians turned a town hall into a referendum on the local data-center boom, the Lake Tahoe story expanded into Silicon Valley vacation pricing, and a tally of the political costs is now a recurring news beat — not an outlier.

News Ars Technica AI

Pennsylvanians use town hall meeting to rail against data center boom

Pennsylvania residents used a town hall to push back on the data-center build-out, framing it as a public-trust and transparency issue.

Why it matters

Backs the Gallup 70%-opposition number with on-the-ground organizing.
Suggests state-level legislation is the likely next step — siting, taxation, transparency.

infrastructure policy regulation

Source →

News TechCrunch AI

Silicon Valley's vacationland needs a new energy provider just as AI is driving prices up

Lake Tahoe will face higher energy prices as AI compute load drives the regional rate base up — a visible cost story for affluent users.

infrastructure policy

Source →

AI psychosis and autonomous-system failure

A viral post from Mitchell Hashimoto claimed entire companies are operating under AI psychosis — leadership steering on chatbot output rather than judgment. Andon Labs' AI radio-host experiment produced the same failure mode in the open: agents left to run a business break it in observable ways. Both prompt the same uncomfortable question for SMB operators: where is the human floor?

News Hacker News

I believe there are entire companies right now under AI psychosis

Mitchell Hashimoto's post arguing entire companies are now operating under AI-induced overconfidence and decision drift hit 1,430 points on Hacker News.

Why it matters

Names a failure mode practitioners have been quietly tracking — operator overreliance on synthesized output.
Strong signal that the operator-side governance gap is now publicly discussed by senior engineers.

safety products market

Source →

News The Verge AI

AI radio hosts demonstrate why AI can't be trusted alone

Andon Labs ran AI agents as radio-station hosts with no human intervention; the result is a usable case study of compounding agent failure in production.

agents safety

Source →

Papers

1 item

The information-integrity crisis tightens

Paper Hugging Face

PreScam: A Benchmark for Predicting Scam Progression from Early Conversations

Benchmark for predicting whether a multi-turn conversation is progressing toward romance, investment, or related scams — defenders need to act earlier in the trajectory.

Why it matters

Treats scams as multi-turn trajectories, mirroring the safety direction of this week's agent papers.
Practical training signal for any platform that hosts user-to-user chat.

safety evaluation benchmarks

Source → Arc

Also today

News · OpenAI OpenAI and Malta partner to bring ChatGPT Plus to all citizens — OpenAI and Malta will offer ChatGPT Plus and training to all Maltese citizens — a country-level deployment template.
News · The Verge AI Google updates its spam rules to include attempts to 'manipulate' AI — Google's spam policy now explicitly treats AI-search manipulation as a violation — putting prompt-injection-style SEO on notice.
News · TechCrunch AI Runway started by helping filmmakers — now it wants to beat Google at AI — Runway is betting video generation is the path to world models — and pitching itself against Google on that thesis.
News · TechCrunch AI Osaurus brings both local and cloud AI models to your Mac — Osaurus combines local and cloud AI in a single Mac app while keeping memory, files, and tools on-device.
News · OpenAI How business operations teams use Codex — OpenAI's playbook for business-ops use of Codex — initiative briefs, leadership decks, strategy updates.
News · OpenAI How data science teams use Codex — OpenAI's data-science Codex playbook — root-cause briefs, impact readouts, scoped analyses.
News · OpenAI How sales teams use Codex — OpenAI's sales Codex playbook — pipeline briefs, meeting prep, forecast reviews.
News · Hugging Face Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — IBM releases an Apache-2.0 multilingual embedding model with 32K context — credible drop-in for retrieval-heavy stacks.
News · Stratechery 2026.20: Shifting Alliances in a Changing World — Stratechery's weekly roundup focuses on a new kind of computing and the shifting alliances the AI build-out is forcing.
Paper · Hugging Face Long Context Pre-Training with Lighthouse Attention — Training-only hierarchical attention that wraps standard SDPA and can be removed at inference — extreme-sequence pretraining without an architecture change.
Paper · Hugging Face WildTableBench: Benchmarking Multimodal Foundation Models on Table Understanding In the Wild — Benchmark for messy, in-the-wild table images — exposes how thoroughly clean-image evaluations overstate multimodal model capability.
Paper · Hugging Face PanoWorld: Towards Spatial Supersensing in 360° Panorama World — Targets 360-degree spatial understanding as a 'supersensing' regime for MLLMs — relevant to robotics, navigation, and 3D scene perception.
Paper · Hugging Face Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning — Closed-loop verification grounds T2I multi-step reasoning — addresses the planning hallucinations and post-hoc reflection that hobble current approaches.
Paper · Hugging Face Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models — Training-free pace-and-path correction for vision-language-action models addresses dynamics-blindness without retraining the base policy.
Paper · Hugging Face Learning to Communicate Locally for Large-Scale Multi-Agent Pathfinding — Local-only communication for large-scale multi-agent pathfinding — scales coordination without global broadcast.