profile

mAIn Street

mAIn Street #282: Advanced US models will get White House-approved voluntary hacking tests; Congress spent 90% of AI spend on ChatGPT; Mark Cuban to offer free AI bootcamps

Published 6 days ago • 7 min read
The White House tests AI hacking, Congress chooses ChatGPT, and businesses confront the integration work behind AI adoption.
mAIn Street Daily Newsletter. AI news for people who actually have jobs to do.
Tuesday, August 4, 2026
AI’s next phase is being shaped by implementation. Governments are testing what models can break, offices are choosing what to buy, companies are fixing old systems, schools are widening access, and artists are asking to be paid when machines borrow from their work.
Top 5
What matters today
01
The tests will measure whether leading models can help conduct cyberattacks after recent agents escaped controlled environments. The government has not yet explained the scoring system or whether results will become public.
Source: Reuters
02
House records show about $100,580 across 798 ChatGPT transactions during the year ending March 31. The spending offers a rare look at which tools government staff are adopting for research, summaries, constituent work and communications.
Source: TechCrunch
03
The startup maps a company’s existing systems and business processes before building agent-powered workflows around them. Its pitch targets the technical debt and fragmented data that cause promising AI pilots to stall.
Source: TechCrunch
04
The programs give underserved high-school students hands-on work with generative AI, career applications and ethical questions. Meals, technology and transportation help are included at participating sites.
Source: GlobeNewswire
05
The video platform offers per-generation payments and a share of subscription revenue. The early rates are small, and its underlying open models still rely partly on broadly scraped material, leaving the consent problem only partly solved.
Source: The Verge
AI Workflows
Branch the chat before the chat becomes the project
ChatGPT can carry the useful history into a clean conversation when your next idea needs room of its own.
A productive ChatGPT session can slowly turn into a junk drawer. The original assignment is still in there, along with revisions, rejected ideas, side questions, pasted research, troubleshooting notes and a few decisions nobody remembers making. The model can keep working, though each new request must compete with more history.
ChatGPT calls the escape hatch Branch in new chat. On the web version, hover over the message where the new direction should begin, click the three-dot More actions menu, then choose Branch in new chat. ChatGPT opens a separate conversation containing the discussion up to that point. The original chat stays untouched, so both paths remain available.
Why branching helps
The new chat begins with the context that still matters and leaves the later clutter behind. That gives the model a clearer assignment, protects the working version in the original thread and makes competing directions easier to compare. A branch also reduces the temptation to keep repairing a conversation that has drifted far beyond its first purpose.
Where it earns its keep
Drafting: Keep the approved facts and strategy, then branch before testing a warmer opening, a sharper argument or a different audience.
Long projects: Branch after a milestone is approved. The new thread can handle the next phase while the original preserves the decisions, files and history that produced the approved version.
Troubleshooting: Keep the main repair plan intact, then branch to test one suspected cause or an alternate implementation without burying the original diagnosis.
Research and decisions: Branch from the shared evidence to examine a skeptical case, a different budget, another policy option or a second recommendation.
Team projects: A teammate can branch a shared project chat to test a new hypothesis while the original chain remains available to everyone. The branch appears beside the original project conversation.
A simple trigger works well: branch when you catch yourself saying, “Keep everything we have done, except now let’s try something completely different.” That sentence usually means the project needs a second lane.
AI Tool Spotlight
yapyap keeps meeting recordings on your own computer
yapyap records meetings and interviews on macOS, Windows and Linux, identifies speakers, and produces transcripts, decisions and action items when recording stops. Its “lenses” can turn the same conversation into different write-ups for different jobs.
The recording stays on the user’s device unless they choose to export it. That makes the tool worth a look for interviews, internal discussions and client calls where privacy matters as much as convenience.
Source: yapyap
More worth knowing
Headlines
Work, business and adoption
TechCrunch
House records show about $100,580 across 798 ChatGPT transactions during the year ending March 31. The spending offers a rare look at which tools government staff are adopting for research, summaries, constituent work and communications.
TechCrunch
The startup maps a company’s existing systems and business processes before building agent-powered workflows around them. Its pitch targets the technical debt and fragmented data that cause promising AI pilots to stall.
Reuters
Investors rewarded strong cloud results and new chip and infrastructure deals. The milestone shows how closely the AI boom is now tied to the value of the largest public companies.
GlobeNewswire
CORE gives small and midsize businesses training, guardrails, automation and ongoing IT support under one service. The launch reflects growing demand for help that continues after the first AI pilot.
TechCrunch
Updated terms describe a service that records meetings and produces summaries, insights and next steps. The privacy-policy changes give customers an early reason to review how meeting data will be handled.
PR Newswire
The platform analyzes quizzes and learning activity, then suggests where a team needs support or where leaders can improve sales, retention and marketing. Its pre-beta program opened August 3.
Business Wire
The company also reported that its measure of borrower default risk remained well above its long-run baseline. AI lending can automate approvals at scale while still carrying the credit risks of the wider economy.
Education, health and public life
GlobeNewswire
The programs give underserved high-school students hands-on work with generative AI, career applications and ethical questions. Meals, technology and transportation help are included at participating sites.
GlobeNewswire
The renamed Houston company plans to connect clinical, molecular and imaging data to support decisions for brain and central nervous system cancers. Its new name and Nasdaq ticker took effect August 3.
Safety, trust and governance
Reuters
The tests will measure whether leading models can help conduct cyberattacks after recent agents escaped controlled environments. The government has not yet explained the scoring system or whether results will become public.
TechCrunch
The cybersecurity company uses AI to probe networks for weaknesses under controlled conditions. Its $250 million funding round arrived days after major model developers disclosed agents breaching outside systems.
Business Wire
The vendor-neutral initiative will develop reference designs, board briefings, tabletop exercises and maturity measures for autonomous systems. The work focuses on resilience after an agent or model causes damage.
Business Wire
The products watch how AI tools and user accounts behave across an organization, then flag unusual access or synthetic identities. They became available August 3.
TechCrunch
Researchers found apps turning televisions into residential proxy nodes that outsiders could rent. AI companies increasingly use these networks to scrape the web, which makes the hidden consumer cost part of the AI data supply chain.
Consumers, creativity and new tools
The Verge
The video platform offers per-generation payments and a share of subscription revenue. The early rates are small, and its underlying open models still rely partly on broadly scraped material, leaving the consent problem only partly solved.
PR Newswire
Brands can test products and messages against more than 1,000 synthetic respondents representing breakfast buyers. The company reports strong correlations with live panels, though customers still need human research for consequential decisions.
AirJelly
The desktop tool organizes activity from the user’s computer into a searchable work memory and proposes timely reminders. Its local-first approach targets professionals who want context without sending every activity stream to a cloud service.
Ctruh
Teams can create virtual showrooms and product spaces for the web and extended-reality devices. It gives smaller marketing teams access to a format that usually requires specialized development work.
The Verge
The comments landed during a wider dispute over generative music and creator rights. Musicians heard a technology argument that reduced collaboration and craft to training data.
The Verge
Their plan would require chart-eligible songs to be substantially human-made and clearly labeled when AI played a role. The proposal puts commercial consequences behind the industry’s authenticity debate.
Global headlines
Reuters
The 2.8-trillion-parameter model climbed text and visual leaderboards. The release adds pressure on U.S. companies competing with fast-improving open and low-cost alternatives.
Reuters
Artificial Analysis estimated that the model costs far less to run than leading U.S. rivals. Falling inference prices can reshape which models businesses can afford to use at scale.
Reuters
The agency is reviewing recent disclosures from OpenAI and Anthropic. The incidents are pushing voluntary safety arrangements toward a more public test of credibility.
Reuters
The government still favors a lighter approach than the European Union. Recent agent breaches are raising the cost of relying on company cooperation alone.
Reuters
The central bank expects near-term demand for chips, power and construction to lift prices before productivity gains arrive. AI spending is becoming a macroeconomic force far beyond technology companies.
mAIn Street is built for nontechnical readers who want the signal, not the sludge.

​Read All mAIn Street Back Issues Here​

mAIn Street

Aric Mitchell

Check out the resources I offer below and sign up for my new newsletter!

Read more from mAIn Street

SpaceX and Tesla picked Texas for a $16.8 billion chip complex, Google Maps learned to book and order, and U.S. productivity beat forecasts. Friday, August 7, 2026 Thursday's strongest AI stories showed up in factories, maps, budgets and classrooms. SpaceX and Tesla committed $16.8 billion to a Texas chip complex, Google Maps added actions people can take from a search, worker productivity beat forecasts, Alphabet prepared another large bond sale, and an AI-centered private-school network...

3 days ago • 9 min read

AI uncovered health risks in sleep studies, Safeway entered ChatGPT, and Zoox put a price on steering-wheel-free rides. Thursday, August 6, 2026 AI moved into ordinary routines and essential systems today. Researchers pulled new risk signals from sleep studies, large U.S. companies organized around infrastructure threats, Safeway connected grocery carts to ChatGPT, Zoox set prices for steering-wheel-free rides, and Google reorganized its AI leadership as core Gemini builders left. Top 5 What...

4 days ago • 10 min read

A court lifted the block on Perplexity's Amazon shopping agent, OpenAI settled a U.S. hiring case, and Texas paused data-center hookups. Wednesday, August 5, 2026 AI collided with the law and public trust today. A federal appeals court lifted the block on Perplexity's Amazon shopping agent, OpenAI settled a U.S. hiring-discrimination case, Texas paused new data-center connections, medical researchers found a trust trap, and customers demanded human backup. Top 5 What matters today 01 A...

5 days ago • 9 min read