Top 10 AI News This Week — September 28–October 4, 2026

This was the week the rules actually arrived. On a single day, September 30, California signed the first state law in the country to write generative-AI procurement standards and a training mandate into public higher education, vetoed the bill that would have given unions a say before a district buys AI that does its employees' work, and declared that AI agents are not members of the public. Meanwhile Microsoft switched two third-party models on by default inside Word and Excel, and Google and OpenAI both started the clock on retiring the custom tools faculty built last year. Here's what community college educators should know.

EdTech & Campus AI

AB 2392 is law: California now sets GenAI procurement standards and a training mandate for every community college

Governor Newsom approved AB 2392 (Fong) on September 30, 2026, and it was chaptered the same day as Chapter 855, Statutes of 2026. It takes effect January 1, 2027. The bill adds Article 14 to the Education Code and carries a title that says exactly what it does: "Public postsecondary education: generative artificial intelligence systems: procurement standards: training." The CCC Chancellor's Office and CSU are required to convene an intersegmental working group on procurement standards and training; UC is only requested to participate. Recommendations are due January 1, 2028. It passed without a single no vote recorded at any stage: Assembly 77-0, Senate 38-0, concurrence 79-0.

The procurement minimums are specific. A system must undergo risk assessment for harms, misuse and bias; must not produce harmful or illegal output; must avoid "excessive sycophancy"; must have safety monitoring with human escalation protocols for self-harm or harm to others; must support transparent incident reporting; must come with written data use, retention and deletion policies and data minimization with regular audits; and must come with full disclosure of all training data sources, including whether copyrighted content or student-generated content was used. Vendors must also be vetted on privacy, civil liberties and labor practices. The statutory definition of sycophancy is real and worth reading once: "the tendency of a GenAI system to overly agree with, validate, or flatter a user even when the user is communicating factually inaccurate or illegal ideas or actions for the primary purpose of optimizing engagement." Separately, the law requires training for students, faculty and staff on systemwide policy, academic integrity, privacy disclosures and system limitations, with annual content review and retained completion records, and a written report to the Legislature within 60 days of executing a systemwide GenAI contract.

A correction to last week's digest. We reported that AB 2392 was two days from becoming law without a signature and warned readers there would be no signing statement and no press release if it enacted by inaction. We were wrong about the mechanism. Newsom signed it, and named it among 13 AI bills in a Sept 30 release. We also flagged AB 2393 as a one-digit trap on the September 27 signed list; AB 2393 (Addis) is in fact "False imprisonment: false arrest: remedies," unrelated to AI, and we could not re-confirm that it appeared on the September 27 list either, so treat that detail as unverified. One more honest flag: a secondary analysis puts the training-records retention period at four years, and we could not verify that number against the chaptered text, because leginfo blocks automated access. The retention duty is in the law; the number is not confirmed here.

For community colleges, this is the first compliance obligation of the AI era that names you directly, and it arrived with no appropriation attached. The training mandate and the records requirement are the parts that reach individual faculty; the training-data disclosure requirement is the part that will reshape what your district can legally buy. Nothing from the Chancellor's Office yet: as of October 5 there is no AB 2392 implementation guidance posted on ai.cccco.edu. Read more →

The rest of September 30: a union notice bill vetoed, and AI agents declared not to be "persons"

AB 2392 did not arrive alone. September 30 was the constitutional bill-action deadline and the busiest single day of the session, with 147 bills signed and 32 vetoed.

AB 2656 (Petrie-Norris) was vetoed. It would have required a public agency to give written notice to the union 45 days before purchasing or developing generative AI that would perform work currently done by public employees. Its death matters for a plain reason: community college districts are public employers, and a 45-day pre-procurement notice duty would have landed squarely on district AI purchasing. With it gone, AB 2392's procurement standards, not a bargaining trigger, are the operative constraint on CCC GenAI buys. No veto message text was retrievable, and the coverage describes the bill's reach as "state and local agencies" without naming community college districts specifically.

SB 1159 (Cabaldon) was signed as Chapter 863. It amends the Brown Act, Bagley-Keene and the California Public Records Act to clarify that "person" and "member of the public" do not include artificial intelligence systems, autonomous agents, or robots, and it permits a public agency to use a disclosure verification tool to detect AI in submissions. That reaches district board governance directly: AI-generated public comment at board meetings and bot-generated records requests are now something a district may screen for. Note what it is not — a narrowing bill, not a transparency mandate. It imposes no AI inventory or disclosure duty on the agency itself.

Also signed the same day, and relevant to colleges as employers rather than as educators: SB 947 (McNerney) on employment automated decision systems, the "robo bosses" bill with a human-corroboration requirement, which is Newsom reversing his own 2025 veto of SB 7; SB 951 (Reyes) on technological displacement notice; and AB 1331 and AB 1883 on workplace surveillance, the latter reaching AI neural data. We did not verify chapter numbers or public-employer applicability for those four, so confirm with counsel before acting on specifics. If you sit on a governance committee or an academic senate, SB 1159 is the one to raise this month, because it changes what your board may do with public comment. Read more →

Florida's college AI rule passed three weeks ago and still has not been filed

We have tracked Rule 6A-14.0719 for three weeks, and this week produced a hard answer. As of October 5, the Florida Administrative Code gateway lists the rule's status as "Not Adopted by FAC Yet" with the effective-date field blank. The State Board of Education approved it on September 16, but board approval is not filing, and nothing has been filed with the Department of State.

The rule's entire public record is two notices: a development notice published May 27, 2026 and a Notice of Proposed Rule published August 25, 2026. Public comment closed September 15. Under s. 120.54(3)(e)2, Florida Statutes, a rule must be filed for adoption not less than 28 days and not more than 90 days after the notice of proposed rule, and a filed rule becomes effective 20 days after filing unless a later date is specified. From the August 25 notice, that puts the filing window at roughly September 22 through November 23, 2026 — so the rule is inside its window, not past it, but the clock is running.

The substance, once it is in force, requires each of the 28 Florida College System boards of trustees to adopt policies governing AI use by students, staff and visitors. And to kill a number that keeps circulating: the July 1, 2027 date in coverage is the K-12 districts' date under Rule 6A-1.0957, not this rule's. The college rule currently has no institutional policy adoption deadline and no effective date at all.

The comparison is the point. Two states moved on college AI policy in the same three weeks by completely different routes — California by statute, with a chapter number, an effective date and a vote count, and Florida by agency rulemaking that is now in its third week of administrative limbo. If your district is waiting to see what a state mandate looks like before writing policy, California's is the one that exists. Read more →

Three community college systems moved on AI this week, and not one of them launched a credential

The Manufacturing Institute, the workforce arm of the National Association of Manufacturers, signed an MOU with the Alabama Community College System on September 29 to deploy its AI Skills for Manufacturing Training Suite free and self-paced across all 24 ACCS colleges. The ceremony was at Calhoun Community College's Gene Haas Workforce Training Center. The suite includes "AI Fundamentals for Manufacturing" plus an "AI Fundamentals+" course launching this fall, and it is funded out of $10 million that Google.org gave the Manufacturing Institute in April 2026. The stated target is 40,000 current and future manufacturing workers over two years. Keith Phillips, ACCS Deputy Chancellor of Workforce and Economic Development: "Adopting the MI's AI Skills for Manufacturing Training Suite ensures our curriculum moves at the speed of modern industry." Alabama is stop one of a three-state roadshow; North Carolina and Texas follow.

In North Carolina, Forsyth Technical Community College landed $916,000 from the National Science Foundation to expand its AI certificate and degree and build out a lab — two robotic arms now, humanoid robots planned in years two and three, with coursework in machine learning, neural networks and visual computing. Nearly 80 students are in its online AI courses this fall. The better detail is older: Forsyth Tech led the consortium that created 24 new AI courses now available across all 58 North Carolina community colleges, and is building a cross-industry AI robotics degree with Central Piedmont. And in Louisiana, Louisiana Delta Community College in Monroe built two four-week credentials with Meta — one for data center technicians, one for construction skills — reporting 100+ graduates since January 2026, all employed, against a wage contrast its director named plainly: nursing assistant work at $16.25 an hour versus fiber installation starting at $20 or more. Meta has promised $5 million in scholarships in the region.

Honest flags on all three. The Manufacturing Institute's "82% of manufacturers report AI skills gaps" figure comes with no sample disclosed, 40,000 is a target rather than an outcome, and ACCS published no release of its own. Forsyth Tech is single-sourced to one local broadcast report with no college release and no NSF award number we could confirm. Louisiana Delta's placement rate and the $5 million are employer- and college-supplied and unaudited, and the AI connection there is indirect — AI-driven data center construction, not AI instruction.

Here is the pattern worth naming, because it is now six weeks old: no community college anywhere launched a new AI degree, certificate or course in this window. What arrived instead was a system adopting someone else's curriculum, a grant expanding a program that already existed, and credentials built for an employer's construction schedule. If your district is waiting for a model to copy, the Alabama MOU is the most copyable thing on this list, and the question for the CCCCO is whether the Manufacturing Institute will do the same deal in California. Read more →

Creative tools, with the costs printed — including one price that doubles on October 8

Four things a design, media arts or communications instructor can use right now, with the numbers attached.

FLUX 3 Image from Black Forest Labs shipped October 1 with region-based editing on a 0–1000 canvas, up to 10 reference images, and native 2K/4K output. The cost matters and it is time-limited: on fal.ai a 1K image is currently $0.024, which is 50% off, and the discount ends October 8 — after which it is $0.048. If you are costing out a fall assignment, cost it at the higher number.

Curious Refuge ran an actual cost test on AI sound design on October 1 and published the figures, which is rarer than it should be: InVideo at $4.56 per project over about two hours, ChatGPT Astra at $5.20 per project over about an hour and three quarters. The verdict is the useful part. Neither produced professional-quality sound design; human-designed audio beat both on timing, layering and polish. Both work as a rough pass a human then refines, which is a defensible thing to teach and an indefensible thing to grade as finished work.

Two free options worth knowing. Depth Anything is open source, runs locally and has a downloadable Mac app, so a depth-map unit costs nothing per render rather than paying per online generation. And TrueScribe does Whisper transcription entirely on-device, with a free tier that already exports SRT and WebVTT — the caption formats Canvas wants — with no account required and nothing leaving the machine, which is the right answer for anything FERPA touches. Its paid tier is $64 one time for three devices, not a subscription. The catch: Windows and Linux only, no macOS.

If you teach graphic design, animation, film or audio, the sound-design result is the one to put in front of students, because it gives them a number for what "good enough to refine" costs and an honest ceiling on what the tools do. Read more →

Major AI Companies

Microsoft turned on GPT-6.1 Sol and Claude Sonnet 5.5 inside Copilot by default

On September 30 Microsoft made OpenAI's GPT-6.1 Sol and Anthropic's Claude Sonnet 5.5 available in Microsoft Copilot, and the Message Center notice (MC1483844) is unambiguous about how: "The feature is enabled by default for eligible licensed users who meet the applicable configuration requirements," and "no action is necessary" to enable access. Rollout began September 30 in Cowork and Copilot Studio, with Word, Excel, PowerPoint and Chat phased in within the week. Eligibility is a Microsoft 365 Copilot user subscription license.

There is one admin catch, and it is the whole story: Claude Sonnet 5.5 requires the Anthropic subprocessor to be enabled, and GPT-6.1 Sol requires the OpenAI subprocessor enabled for broader availability. That is precisely the setting a data-protection review would have flagged — a third-party model becoming selectable in Office apps for licensed users with no admin opt-in, gated only by a subprocessor toggle. Microsoft names no education SKU in the announcement: not A1, A3 or A5, not M365 Copilot for Education, not students. Any district holding Copilot licenses should assume it applies and verify rather than wait to be told.

Two concrete actions. For IT: review the Anthropic and OpenAI subprocessor state against your student-data agreements before the phased app rollout finishes, and make that a deliberate decision rather than a default one. For faculty: the model picker in Word and Excel now offers models your syllabus AI policy probably does not name, which is a one-line fix worth making before midterms. Read more →

Google ships a student hub to Education users of every age, and starts the clock on retiring Gems

Two Workspace changes, one free and one not. On October 1 Google launched a student hub in Gemini, consolidating study notebooks, flashcards and practice quizzes in one place. The availability line is the news: it is "available to Google Workspace for Education users of all ages," not gated to a specific Education edition and not age-gated, which makes it the first of this study-tools cluster to reach under-18 Education users. It arrived October 1 to both Rapid Release and Scheduled Release domains, so a district on Scheduled Release got it at the same time as everyone else. There is no separate end-user setting — the Gemini app toggle in the Admin console is the only control. Students reach it at gemini.google.com/students with nothing to configure and nothing for an instructor to set up.

The integrity exposure is specific rather than general: auto-generated flashcards and practice quizzes built from uploaded sources means a student can convert your course packet, or an old exam bank, into study artifacts inside your own tenant. Worth knowing the toggle state for your student OUs before someone else discovers it for you.

The second change costs money. On September 30 Google announced Skills — reusable, stackable custom instructions built on the Markdown SKILL.md standard — and with it, a dated retirement path for Gems: Skills rollout began October 5 in Workspace and October 13 in the Gemini app; Gems move to the Settings panel November 17; Gems are disabled for business and enterprise no sooner than March 1, 2027, and no sooner than June 1, 2027 for Education users, with 30 days' notice before auto-migration. Education gets three months longer than business. Two catches worth forwarding: Skills in Workspace for Education requires the paid Google AI Pro for Education add-on, unlike the free student hub, and Workspace users must be 18+, so dual-enrollment high schoolers in your tenant cannot use Skills even where their adult classmates can. Any faculty member who built a Gem for a course should know that auto-migration produces a draft skill, not a live one — someone has to publish it. Read more →

OpenAI's custom GPT migration is still conditional with 67 days to go — and the week's model prices all landed at $2/$10

We said last week we would confirm whether custom GPT migration actually reached workspaces by its October 1 second target. It has not, and the answer is more awkward than that. The retirement FAQ, last updated around October 3, still reads conditionally: "When migration is available for your account, go to My GPTs and select Migrate to plugin," and "a target migration date does not guarantee that every account can migrate on that day." The FAQ's own stated target is September 22 for Enterprise workspaces. The October 1 date is not in OpenAI's text at all — it traces to secondary coverage and applies only to Enterprise workspaces that had already opted into delayed banners. And OpenAI's help center contradicts itself: the "Creating a GPT" article gives the migration target as September 17 while the retirement FAQ says September 22.

The retirement dates have not moved and they are the ones that matter: October 26 creation of new custom GPTs ends; December 11 custom GPTs stop working; February 11, 2027 only for workspaces with an approved deferral. The FAQ contains no ChatGPT Edu-specific guidance at all. For a campus running Edu, that is the story: 67 days to December 11, no documented Edu path, no guaranteed migration date, and deferral requires an approved request. Inventory faculty-built GPTs now, check My GPTs weekly for the Migrate button, and file a deferral if a course depends on one past December 11.

On pricing, the week produced a coincidence worth noting and one trap. OpenAI's GPT-6.1 Sol (September 29) is $2.00 per million input and $10.00 per million output, and ChatGPT Edu is explicitly named among the plans getting it in ChatGPT Work and Codex. Anthropic's Claude Sonnet 5.5 (September 28) landed at the identical $2.00/$10.00 — but read that carefully, because the per-token rate is unchanged from Sonnet 5. Anthropic's claim is up to 30% less per task through efficiency, not a rate cut; do not budget it as one. The trap is Google's Gemini 4 Argon (September 30), whose $2/$10 is an introductory price that becomes $4/$20 — a 100% increase — afterward, and which no college can get today anyway, since it is limited to trusted cyber defenders under Google's Fairwind Program. If a campus project is budgeting against frontier Gemini, budget $4/$20. One more for IT: OpenAI's new always-on "dots" agents are off by default for Edu workspaces and require an admin to enable the beta. Leave them off until your integrity policy names autonomous agents, because a persistent agent that works unattended on a schedule is a different problem from a chatbot. Read more →

Free & Student Opportunities

Free credentials, priced honestly — starting with one the CCC already pays for

The best free AI professional development for a California community college instructor is the one the system already runs. "AI Literacy for Educators: Supporting Student Success" is free, self-paced, with badging available, delivered on CVC@ONE rather than the Vision Resource Center, and open to CCC faculty, instructional designers and academic leaders with "access for all interested." No cohort deadline. Duration is not published.

Beyond that, the honest ledger. Anthropic's AI Fluency courses are genuinely free on Skilljar, require a Skilljar account but not an Anthropic or Claude account, and issue a certificate of completion after a final assessment. Two to know: AI Fluency for Educators, and AI Fluency for Creative Work, which is aimed at design, media arts and marketing. The catch is now confirmed on the course pages themselves — both assume the 4D framework from the separate "Framework & Foundations" core course, so this is a two-course sequence, not a one-off. OpenAI Academy courses are free, 30–180 minutes each, with badges issued through Accredible behind an 80% assessment gate — but two corrections to what gets written about them: a ChatGPT account is required to enroll and track progress, and OpenAI states plainly that the badges "are not certifications." Do not let a student put one on a resume as one. Google's AI Educator Series is genuinely free with an ISTE-aligned digital badge after a short quiz, 30–45 minutes for the higher-ed sessions; its next module lands October 7.

And the standing corrections, both of which survived another week. Google AI Essentials is still not free — $49 a month on Coursera after a 7-day trial, re-confirmed on grow.google, and still described as free in education newsletters. Newer and sneakier: OpenAI's "ChatGPT Foundations for Teachers" on Coursera is genuinely free, about five hours, and awards nothing — the course page states "no certificates, credentials, or reports are awarded." Free and useless as a resume line are different problems, and that one is both. AWS Educate remains free with email only, no credit card, age 13+, with digital badges — but its five published paths are storage, compute, networking, databases and cloud operations, with no dedicated AI track, so pitch it as cloud foundations. We could not verify current NVIDIA DLI course terms from a first-party catalog page this week and are not printing a cost claim for it.

The counterexample is worth one line, because the headline will be reprinted as an opportunity. Anthropic announced Claude Frontier Academy on October 2 — $100 million to train 10,000 engineers by the end of 2027, with a "Claude Frontier Deployed Engineer" badge expected in early 2027. Entry is by nomination only, through an Anthropic account team, and the initial cohorts are eight consulting, banking and pharmaceutical firms. There is no community college on the list and no open application path. A $100 million AI training program with no higher-education on-ramp is a fact about the market, not an opportunity. Read more →

October scholarship check: one usable AI scholarship, and it is a transfer play

We check AI scholarships on the first run of each month, and October's sweep came up thin enough that saying so is the finding.

One cleared the bar. The DoD SMART Scholarship-for-Service Program closes December 4, 2026 at 5:00 p.m. ET. It pays full tuition plus a $30,000–$46,000 annual stipend, a summer internship and guaranteed post-graduation employment, in exchange for a 1:1 civilian service commitment. Eligible disciplines include computer science and engineering, data science and analytics, and software engineering. Requirements: 3.0 GPA, citizenship of the US, Australia, Canada, New Zealand or the UK, and the ability to obtain a security clearance. The catch for this audience is decisive: associate-degree students are not eligible as such. This is a transfer play — pitch it to students who will be enrolled in a bachelor's program, not to a first-year student. Apply at smartscholarship.org. One disclosure: that site is JavaScript-rendered and would not load for us, so the deadline is confirmed on the administering organization's page rather than the program's own.

What did not clear the bar, and why it is worth printing: the AFCEA Nightwing Scholarship is a $5,000 award that explicitly lists artificial intelligence as an eligible field and explicitly includes community college students — the single best structural fit we found — but its page reads "Applications closed!" with no future date posted. AWS AI & ML Scholars closed June 24 with no 2027 date announced. Generation Google Scholarship could only be verified back to a 2024 cycle. Everything else that surfaced was a scholarship directory or lead-generation site, which we do not run. If you advise students, the practical move this month is to set a reminder to check AFCEA Nightwing rather than to chase the open ones. Read more →

AI Competitions

Closing soon, and all of it real cash

Five open hackathons with verified deadlines and cash that is actually cash. Deadlines are exact, and two of them will trip you on time zone.

Build With AI: Basics (Devpost's own) — October 26, 5:00 p.m. EDT. $2,500, 100% cash ($1,250/$750/$500). Ages 18+, no student restriction, so faculty and staff can enter; dual-enrolled high schoolers cannot. Free, beginner-friendly, and the lowest-barrier real-cash item on the list. learn-ai-basics.devpost.com

IEEE ClimateChain Global Hackathon — October 25, 5:00 p.m. MSK, which is 4:00 a.m. Pacific on October 25. Read that twice; a student reading "5:00 p.m." will miss by a day. $3,000, all cash ($1,500/$1,000/$500), plus Certificates of Merit and IEEE Senior Member nomination eligibility. Students only, above the age of majority, companies excluded, free, 539 registrants. Requires a working prototype, a 3–5 minute demo video and a public repo. Same organizer family and the same prize shape as the OneAquaHealth hackathon we ran, which means a small field. ieee-climatechain-hack.devpost.com

Amazon Build, Ship, Shape — October 23, 12:00 p.m. PDT — noon, not midnight. $138,000, all cash, with AWS credits added on top rather than substituted. Ages 18+, no student bar, free, no hardware purchase required (Fire TV, Bee and Ring all have simulators). Brazil, Quebec, Russia, Crimea, Cuba, Iran and North Korea are excluded, though that list is in the full rules rather than the sidebar. Deep platform lock-in per track. amazonappdev2026.devpost.com

OpenCV AI Competition 2026 — October 26, 11:45 p.m. PDT. Advertised at $20,250. The real cash is $12,000 ($5,000/$3,000/$2,000 plus two $1,000 specials); the other $8,250 is 55 AWS compute grants of $150 each, and Devpost's prize template still mis-renders those grants as "in cash." 18+, no student restriction, free. opencv26.devpost.com   Nebius x NVIDIA Global AI Hackathon — October 30, 10:00 a.m. PDT. $50,000, all cash, plus four Jetson Orin Nano units; 18+, no student bar. Requires Nebius infrastructure and an NVIDIA open-source model — double lock-in. nebiusglobalaihackathon.devpost.com

And a correction readers need before they waste an afternoon. The OneAquaHealth IEEE hackathon we recommended closed on October 4 at 9:00 p.m. PDT and was not extended. Its page still displays "Registration Still Open" with a live Join button, and it still appears in Devpost's own open-hackathons API with an end date rendered as October 5. Neither means you can enter. That also revises our own verification method: appearing in the open API is necessary but not sufficient, and the sidebar deadline has to be checked against the clock. One more with a caveat rather than a recommendation: the ASU North American AI Challenge closes October 7 with $22,500 all cash, but it requires pre-made teams of 3–5 students, publishes no deadline time zone, and does not state whether two-year colleges count as "eligible North American universities." A Fullerton team should email before investing two days in it. Read more →

The gap between advertised prizes and real money, in two spectacular examples

This section exists partly to keep students from wasting a weekend, so here is the week's evidence. ForgeHacks Online 2026 advertises $1,097,385 in prizes. The actual cash is $291 — $100 for first, $50 for second, $25 for third, $100 for the AI and cybersecurity track, and roughly $16 spread across participant perks. Everything else, about $1,097,094 of it, is credits, subscriptions and software licenses, and the page itself carries repeated "THIS IS NOT A CASH PRIZE" markers, including one that states outright "Cash portion is $100." That is a ratio of roughly 3,770 to 1 between advertised value and money. Students only, closes October 10 at 12:00 p.m. EDT.

ML Empowerment Build Challenge 3.0 puts "$400,000 in prizes!" in its title. The actual cash is $0. Devpost's API reports the prize amount as zero and the sidebar labels all three prizes non-cash; the pool is platform credits and subscriptions, inflated chiefly by a $500 credit handed to all ~620 participants. Ages 13+, students only, closes October 9. And UnivaBio, which we flagged twice before, did not close October 6 — it was extended to October 13, still with $0 cash against roughly $40,000 advertised, most of it 100 winners receiving $299 in platform credits. Correct the record from "closing" to "extended, still worthless as cash." Galuxium Nexus V2 is still running through October 31 and still disqualified here for the same reason.

Two forward-looking notes instead of more warnings. The Community College Innovation Challenge, run by AACC with NSF, is the single best structural fit for this audience that exists: teams of two to four students "seeking a two-year degree in the U.S. or its territories," with a faculty or administrator mentor required and all members 18+. The 2026 cycle closed April 3; expect the 2027 window to open in January, and it is worth putting on a department calendar now. And the Presidential AI Challenge, which explicitly includes educators as well as students, says to check back in mid-October — so we will. Read more →

Embodied AI & Robotics

California's new robotaxi law invents a job, and Minneapolis spent the week arguing about the ones AI takes

Newsom signed SB 1246 (Cortese) on or about September 30, and it is the rare AV bill that creates headcount rather than just rules. Its provisions phase in by July 1, 2028. Remote operators must be U.S.-based and hold valid U.S. driver's licenses; AV companies must notify local jurisdictions of vehicle locations during system-wide failures; they must field on-site "local incident technicians"; local penalties attach when an AV obstructs first-responder operations for 30 or more minutes; and reporting on company response times expands. The California DMV will write the actual response-time standards. Cortese: "California has embraced autonomous vehicles, but we cannot embrace innovation at the expense of public safety." Waymo said it will comply.

One number to refuse: the $10,000 fine figure in several headlines is not in the senator's release and traces to July bill-stage reporting. We are not printing it as enacted, and neither should anyone else without the bill text in hand.

The same week, the Minneapolis city council passed a requirement for human safety monitors — licensed and paid at least minimum wage — in autonomous rideshare vehicles, 7–6, with the mayor pledging a veto and seven votes short of an override, so it is very likely dead. It belongs here anyway because of what the debate was about. Both supporting council members argued workforce, not safety: one citing immigrant drivers for whom automation "is an existential threat," the other that "automation has a demonstrated track record of leading to job displacement."

For community colleges, the honest framing is that SB 1246 creates a plausible new California job category on a 2028 clock — local incident technician, plus U.S.-based licensed remote operators — and that no college, program or training pathway is named anywhere in the bill or the release. The pathway is inferred, not announced. But the DMV rulemaking that will define the role does not exist yet, which is exactly the window in which an automotive technology or public-safety program with something to say could say it. Note also that no new California community college robotics, mechatronics or AV-technician program, lab or partnership was announced this week; we looked five ways. Read more →

This roundup was assembled with AI-assisted research by the EDUAIATLAS team. Spot something we missed? Let us know.