Sponsored by
Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking (relatively) ancient programming languages, AI-powered pentesting, and more!
AI in Action: Taking on legacy code
OpenAI claims that its newest models have made significant strides in writing and migrating code in COBOL, Fortran, and Visual Basic. The company calls these “low-resource” languages because far less public code exists to train on. That matters because IBM estimates COBOL still runs more than 40 percent of online banking systems and 95 percent of ATM transactions. On OpenAI’s new COBOL migration test, which asks a model to translate programs into C++, Java, Kotlin, Python, or Rust, GPT-6.1 Sol scored 86.6 percent at low reasoning effort, up from 5.4 percent for GPT-5.6 Sol. At higher effort, it reached 97.9 percent.
The broader benchmark, MultiLangBench, asks models to build a SQLite-compatible database engine from scratch in 20 languages. GPT-6 Astra averaged an 87.4 percent pass rate, compared with about 38 percent for GPT-5.6 Terra. COBOL stayed the hardest language, though it climbed from 10.9 percent to 72.1 percent, while Fortran rose from 21.4 to 83.5 percent and Visual Basic from 27.5 to 89.4. Fortran is hardly a museum piece, either: OpenAI says it’s used in weather forecasting and nuclear reactor safety analysis. MultiLangBench is going up on GitHub, so outside teams can check the numbers against their own code.
The reason these scores carry weight is how the migration test grades. A translated program has to preserve the original’s behavior exactly, down to financial calculations and output formatting, because changing either could change what a bank owes or reports. That risk is a big part of why institutions have left decades-old systems alone. At low reasoning effort, Sol’s result also came at 61 percent of the API cost of the previous model, so a team can run a translation and compare it against the original program’s output without a huge bill. OpenAI is now asking the people who maintain these systems which languages and workflows it should tackle next.
🔥 Rapid Fire
Commentary: AI Had Better Work Out
Analysis: Venture Capital Is Killing Itself
Amazon is paying billions for GPUs that it can’t install
“Some AWS data center sites had room for more racks but not enough available power to turn them on, leaving the equipment to be staged at intermediary warehouses until the facilities were ready.”
“At least one AWS supply chain team did not have ‘a clear forecasted path’ for getting racks into data centers, according to one of the emails reviewed by Business Insider.”
Nscale data center faces delays ahead of IPO
“The facility is billed as a flagship site for the British data center developer, but construction is six months behind schedule, set back by leaks in the cooling system, local regulations, construction delays, and a fire.”
Nscale is horribly unprofitable per S-1 (page 20)
Investors urge Nscale to trim IPO valuation target
Microsoft withdraws from data center deal with Nscale
HSBC issues CoreWeave warning — cites ‘structural concerns for neoclouds’
AI returns are not large enough to justify neocloud buildout
“At the moment, we have more compute capacity than we have demand, and we have a lot of the tokens and processing power being given away in freemium solutions.”
NextDC shares plunge 44% after data center delay
“NextDC slashed its full-year earnings guidance by a third and placed development plans on hold after protests delayed power supply to one of its major projects in Melbourne.”
Anthropic pursues $60B in debt from Blackstone at a >10% interest rate
Bank of America warns clients AI investment boom resembles past busts
Companies rushed in on AI. Now some are pulling back.
Websites are closing their doors to AI agents
Your Work May Qualify You for New Immigration Options
You’ve built systems, shipped products, and solved hard technical problems. That record may be the foundation of a U.S. Visa or Green Card case.
Manifest Law helps software engineers and technology professionals assess talent-based pathways like the O-1, EB-1, and EB-2 NIW.
Take our five-minute quiz to share your experience and goals, and an immigration professional will help you see which options may fit and what to take on next.
Attorney Advertising
📖 What We’re Reading
An application can pass vulnerability scans and still be exploitable. The reason is simple: not every security flaw sits inside a vulnerable endpoint. Some hide in the way users, roles, transactions, and application states interact. That’s where business logic becomes a serious AppSec problem.
