Today's flow
The cost of handing work to AI is falling while the amount handed over is climbing fast. OpenAI released GPT-6.1 Terra, which reads the screen and operates apps directly, at half the price of its predecessor, and Anthropic's IPO filing shows business customers spending nine times what they spent in the same period a year ago.
The same automation has reached attackers. In a hacking case announced by Korean police, an AI tool found weak spots in small businesses such as golf courses, cram schools, and clinics on its own. On jobs, the Bank of Korea pointed to shrinking demand for entry-level youth work, and Microsoft offered voluntary retirement to up to 60,000 employees. Today is a day to review together what to hand to AI, what to protect, and how to keep growing people.
GPT-6.1 Terra takes on computer work at half the price
OpenAI released its mid-tier model GPT-6.1 Terra on October 26. It is available right away to ChatGPT Plus, Pro, Enterprise, and Edu users and in the coding tool Codex, and developers call it as gpt-6.1-terra through the API (the connection point other programs use to call the AI). API pricing is $1 per million input tokens and $5 per million output tokens. Tokens are the small pieces AI splits text into for counting, and they are the unit used for pricing and usage limits.
OpenAI says this price is 50% lower than the earlier GPT-5.6 Terra and the higher-end GPT-6.1 Sol, and that it finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. The emphasis is on computer use: reading the screen and clicking and typing across multiple apps. The company lists form filling, video editing, reading technical drawings, and repetitive office tasks as uses, and says that in internal evaluations at low reasoning effort (a setting that limits how much the model thinks before answering) the error rate fell from 53.6% to 16.7%.
If people on your team still copy data between screens by hand, now is the time to test it. Lower prices mean more tasks on the same budget, but work where mistakes turn directly into costs should keep a review step. OpenAI still recommends Sol or Astra for the most complex coding.
The speed and accuracy figures come from OpenAI's own evaluations and have not been independently verified. Whether you get the same results on your own forms and screens is something you will only know after running a few real cases.
Anthropic's S-1: business customer spend up 9x in a year
Anthropic made its IPO registration statement, the S-1, public with the US Securities and Exchange Commission (SEC) on October 26. It plans to list on Nasdaq under the ticker ANT, and the price range is still blank. Revenue for the first half of this year was $16.29 billion, about 14 times the $1.16 billion of the first half of last year. In the second quarter it posted its first quarterly operating profit since its founding, at $974 million.
For the first half as a whole, it recorded an operating loss of $693 million. Most of the $80.7 billion net loss, $72.39 billion, is a valuation charge on convertible notes (debt that can later be turned into shares) and similar instruments as the company's value rose, so it differs in nature from losses paid out in cash. Customers spending more than $100,000 a year grew to about 11,900 from about 950 a year earlier.
The most striking figure is net revenue retention of about 900% as of September 30. It means business customers who were already paying a year ago spent nine times as much in the latest three months as in the same three months a year earlier, excluding flat-rate Pro and Max individual subscriptions. It signals that once a company starts using AI, usage grows quickly, so it is wise to set usage caps by department and a cost-reporting cycle from the start.
Net revenue retention is a metric defined by the company, and the filing itself notes that it rests on numerous assumptions. The listing date and offer price have not been set.
AI pentest tool ARTEX used to steal 8.2M records, extort 38 firms
On October 26, the Cyber Investigation Division of the Seoul Metropolitan Police Agency said it had caught a ring that used ARTEX, an AI autonomous penetration testing tool, to steal more than 8.2 million personal records from about 90 Korean businesses. Penetration testing normally means attacking a system with permission to check its security. The ringleader, a 32-year-old identified as A, was sent to prosecutors under arrest, and six accomplices are being investigated without detention. They demanded a total of 360 million won from 38 businesses in cryptocurrency such as Tether, and only one small business actually paid (300,000 won).
The crimes ran from September 2025 to October this year, with ARTEX in use from late August. According to police, once given a target site, the tool picked suitable attack tools on its own and went after weaknesses in databases (the stores that hold customer information) and admin pages. Work that once required the attacker to type each command was done by AI. The damage was concentrated on small businesses: 21 golf courses, 12 cram schools, 5 hospitals, 4 universities, and 3 each in insurance sales and funeral services.
What matters is that the targets were golf courses, cram schools, and clinics rather than large companies. If the admin page of an outsourced website is open to the internet or old software is still running, it will be among the first hit by automated attacks like this. Police urged victims not to negotiate or pay when threatened with leaked data, and to report immediately.
Police said no direct link has been confirmed between this case and recent hacks in the financial sector. They are still tracking unreported victims and criminal proceeds.
Bank of Korea: AI shrinks demand for entry-level youth work
The Bank of Korea published a BOK Issue Note on weak youth employment and policy tasks on October 26. The employment rate for young people aged 15 to 29 was 43.7% from January to September this year, and the share of temporary and daily workers rose from 30.7% in early 2022, just before generative AI arrived, to 32.6%. A job-quality index combining factors such as pay and stability fell for young people from an average of 101.1 in 2014 to 2019 to 94.7 since 2023, while it rose for those aged 30 to 59 and over 60.
The report found that after generative AI spread, jobs and industries highly exposed to AI came to prefer experienced hires, reducing demand for the entry-level and basic work young people used to do. It also estimated that the share of young people moving from small firms and non-regular jobs to large firms and regular jobs fell from around 8% in 2014 to 2019 to around 5% in 2023 to 2024. However, it also cited an industrial structure centered on IT manufacturing, which creates relatively few jobs, as a main background factor.
If your company is moving data sorting and first drafts that used to go to new hires over to AI, decide at the same time where people will get the learning that work used to provide. When entry-level work disappears, the pipeline for growing experienced staff a few years later narrows too.
The report does not single out AI as the sole cause of weak youth employment. The direction the Bank of Korea proposes is to pursue industrial policy and labor market reform together and to align education with labor demand in the AI era.
Microsoft offers voluntary retirement to up to 60,000
Bloomberg reported on October 26 (local time) that Microsoft offered voluntary retirement to up to about 60,000 employees at vice president level and below in the US, Canada, the UK, and India. That is about 27% of its roughly 223,000 employees as of the end of June. Xbox, Windows, the AI service Copilot, and finance are included, while staff in areas such as data centers, sales, marketing, cybersecurity, and quantum computing are excluded.
The 60,000 figure is the number eligible, not the number who will leave. In a memo to staff, Chief People Officer Amy Coleman described the offer as another tool for each business unit to respond to changing workforce needs. Reports cite the sharp rise in spending on AI model development and data center construction as the background. In an earlier offer this year to about 8,750 retirement-age employees, more than 30% accepted.
Including the team that builds the AI assistant Copilot shows that the simple idea that AI-related teams are safe does not hold. Big companies are moving money from people to computing infrastructure, so your own AI adoption plan should also spell out which work will shrink and which skills you will build.
There has been no public announcement from the company, and the actual uptake is not yet known. Microsoft is scheduled to report quarterly earnings on October 28 local time.
Things to try today
Open the admin page address of your company website or booking site using mobile data outside the office. If the login screen appears as is, ask your outsourcing or hosting provider to restrict access by location and turn on two-factor authentication.
Pick one screen task that someone copies by hand and hand it to GPT-6.1 Terra in ChatGPT. Before you do, write down how you will check the result against the original, and if errors turn up, keep a human review step for that task.
Put your team's AI service usage and bills for the past three months into one table and set a monthly cap per department. Today's numbers show usage can multiply within months.
Write a one-line response order for when you receive a threatening email or text over leaked personal data: who reports to the police and who handles customer notices. The first line is not to pay.