Sunday’s 5 PM technology recap covers five developments that matter beyond the headline: OpenAI published internal measurements on coding-agent use, an independent developer documented Chrome retaining Google.com data despite a delete-on-close setting, three Mount Shasta hikers were rescued after relying partly on Gemini for planning, AI companies’ election spending met widening data-center opposition, and UBS reportedly made AI fluency a hiring requirement for future junior bankers.[1][3][4][6][9]
Reporting cutoff: September 6, 2026, 5:15:04 p.m. EDT (21:15:04 UTC). Sunday produced one clearly same-day vendor publication and several useful reports whose events or date-only timestamps sit just outside—or cannot be proven inside—the strict post–Saturday 5 PM window. Those near-window fallbacks are labeled rather than presented as breaking after the cutoff. OpenAI’s productivity figures are internal measurements, the Chrome finding is an independent two-device test without a checked Google response, and the UBS requirement comes from Financial Times reporting.[1][3][9]
Listen to the broadcast-ready recap
This KittenTTS Bella narration was synthesized locally from the same checked facts as the article.
Tonight’s five-story rundown
- OpenAI says coding agents now supply 3.1 agent-workdays per human workday inside its research organization.[1]
- A developer finds Chrome retaining Google.com site data despite delete-on-close.[3]
- A Mount Shasta rescue becomes a warning against AI-only trip planning.[4]
- AI infrastructure becomes a larger election and community-planning issue.[6]
- UBS reportedly moves AI fluency into entry-level hiring.[9]
1. OpenAI says coding agents have become an internal research multiplier
OpenAI published “Research acceleration: The view inside OpenAI” on September 6 and said its measurements show it has reached its previously announced goal of an “automated research intern”: a system that can complete well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days.[1] OpenAI also says people still choose priorities, judge results, and decide whether work should scale, pause, or deploy.[1]
As of mid-August, OpenAI says its research organization used 3.1 agent-workdays for every human workday, treating eight hours of agent runtime as one agent-workday.[1] The company reports that the median researcher used more than $600 per day of inference at API prices, while the 90th-percentile user exceeded $7,000 per day.[1][2]
Practical action: teams adopting coding agents should measure accepted output, test pass rate, defect escape, reviewer time, compute cost, security incidents, and rollback frequency. Parallel activity is useful only when verification keeps pace with generation.
2. Chrome’s delete-on-close setting may not clear Google.com data as users expect
Independent developer Jeff Johnson reported that Chrome 152.0.7977.83 on two Macs retained Google.com cookies, Local Storage, and Session Storage after he enabled the setting to delete site data when all windows close.[3] In his reproduced test, Chrome’s site-data page showed about 1,216 KB remaining for Google.com after the browser window closed and after Chrome was relaunched.[3]
Johnson says he was not signed into Chrome, had disabled Chrome sign-in, changed the default search engine to DuckDuckGo for the test, and did not observe the same persistence for other sites he checked.[3] He also says he does not know when the behavior began and frames it as a possible bug rather than proof of intent.
Practical action: privacy-conscious Chrome users can review chrome://settings/content/all after closing and reopening the browser, manually remove unexpected site data, and avoid treating delete-on-close as a substitute for account controls, private browsing, or a separate browser profile. Managed environments should test the exact Chrome build and policy set they deploy.
3. A Mount Shasta rescue shows why AI cannot be the only safety planner
TechCrunch reported at 12:35 p.m. PDT on September 5 that three hikers were rescued after a Mount Shasta trip planned partly with Google Gemini.[4] According to the cited Siskiyou County Sheriff’s Office report, the group reached the summit around 7 p.m., descended in darkness, spent the night in Mud Creek Canyon, and was rescued the next morning by U.S. Forest Service rangers and volunteers.[4][5]
The sheriff’s office said the hikers had been advised by Gemini to bring far less food and water than the group needed once an expected eight-hour ascent became a multiday ordeal.[4] TechCrunch also cautioned that Gemini cannot fairly take all the blame for the group’s decisions.
Practical action: for hikes, storms, medical decisions, travel restrictions, or emergency procedures, use official local guidance, current conditions, human expertise, maps that work offline, and conservative supplies. Treat AI output as an unverified draft—not the safety authority.
4. AI data centers are becoming an election issue, not just a construction project
The Independent reported on September 5 that AI companies and affiliated political groups had pledged roughly $265 million for the 2026 U.S. elections, citing a Wall Street Journal analysis.[6][8] The report connects the spending to rising local opposition over electricity costs, water use, land, tax incentives, and the distribution of promised jobs.
The $265 million figure describes pledged political spending, not the cost of data centers, completed election spending, or a measured change in voter opinion.[6] The Independent’s report was published at 8:23 a.m. EDT on September 5, before the strict window, and is included as a disclosed near-window development because infrastructure decisions directly affect local businesses and residents.
Practical action: communities evaluating a proposed site should demand independently reviewable load forecasts, water plans, utility-rate protections, tax terms, construction and permanent-job estimates, backup-generation details, noise limits, emergency planning, and enforceable decommissioning obligations.
5. UBS reportedly makes AI fluency part of the entry-level hiring gate
The Financial Times reported on September 5 that UBS will require graduates and interns applying for junior investment-banking roles to demonstrate proficiency with AI tools.[9] The report concerns future applicants and hiring expectations; it does not establish that every UBS role worldwide has the same requirement.
Accessible follow-on reporting says candidates may face AI-related questions and are expected to show how the tools can support analysis, research, and client work.[10] Because the Financial Times page displayed September 5 without a reliably accessible exact publication time in our checks, this is a date-only near-window fallback rather than a claimed post-cutoff story.
Practical action: job seekers should practice using approved tools to research, structure analysis, test assumptions, and document sources, while learning when not to upload confidential material. Employers should define allowed systems, review expectations, retention rules, and escalation paths before testing candidates on AI.
What to do next
- Teams using coding agents: measure accepted work and review cost, not agent runtime alone.
- Chrome users: verify delete-on-close behavior on the exact build and device you use.
- Travelers and hikers: put official local guidance and conservative supplies ahead of chatbot advice.
- Communities: evaluate data-center proposals with enforceable power, water, tax, and job evidence.
- Job seekers and employers: pair AI fluency with confidentiality rules, source checking, and domain judgment.
Need help securing or improving your technology? Schedule The IT Guys
More from The IT Guys
Frequently asked questions
Does OpenAI’s 3.1 ratio mean agents replaced three researchers?
No. OpenAI counted eight hours of agent runtime as an agent-workday. The company does not present that ratio as three completed human jobs or three times the verified output.[1]
Does Chrome always keep Google.com data after closing?
That has not been established. Johnson reproduced the behavior on two Macs running Chrome 152.0.7977.83; other operating systems, builds, profiles, and policies need separate testing.[3]
Should I never use AI for trip planning?
AI can help create questions and checklists, but safety-critical plans should be verified with official local sources, current conditions, qualified people, and offline contingencies.
Is $265 million the amount already spent building data centers?
No. The reported figure concerns pledged election-related spending by AI companies and political groups, not construction costs.[6]
Will every junior banking job require AI proficiency?
No evidence checked here supports that broad claim. The Financial Times report specifically concerns UBS graduates and interns applying for junior investment-banking roles.[9]
Sources checked for this edition
- OpenAI — “Research acceleration: The view inside OpenAI” — September 6, 2026; company publication and internal measurements.
- Unite.AI — “OpenAI Hits Goal of Building an ‘Automated Research Intern’” — September 6, 2026; independent summary of OpenAI’s disclosed figures.
- Jeff Johnson / Lapcat Software — “Chrome again exempts Google from user site data settings” — September 5, 2026; independent reproduction.
- TechCrunch — “Hikers rescued after using Google Gemini for planning” — September 5, 2026, 12:35 p.m. PDT.
- Chicago Tribune — Mount Shasta rescue report — September 3, 2026; underlying report cited by TechCrunch.
- The Independent — “AI companies are dumping $265M into the midterms as data center backlash sweeps through communities” — September 5, 2026, 8:23 a.m. EDT.
- Yahoo News syndication of The Independent report — September 5, 2026, 12:23 p.m. UTC; not independent corroboration.
- The Wall Street Journal — AI lobbying and 2026 elections analysis — source analysis cited by The Independent; paywalled during this check.
- Financial Times — “UBS demands new junior bankers show AI proficiency” — September 5, 2026; page blocked automated full-text retrieval.
- Startup Fortune — “UBS Now Requires New Junior Bankers to Prove They Can Use AI” — September 6, 2026; follow-on summary attributed to the Financial Times.
Source state checked through September 6, 2026, 5:15:04 p.m. EDT / 21:15:04 UTC. Sunday’s strict after-cutoff news was thin, so four clearly disclosed near-window fallbacks were used rather than inventing or padding stories. Search snippets were used only to recover metadata where the source page blocked automated retrieval.