A roundup of what happened in IT during October 26 – November 1, 2026. This follows Week #43 of 2026.
Microsoft Offers Voluntary Exits to 60,000 Employees
On Monday Microsoft sent a Voluntary Exit Offer to roughly 60,000 employees — about a quarter of its workforce — in the US, Canada, India, and the UK, at vice-president level and below. The company expects only 3–5% to accept, fewer than 3,000 people, and employees have until early December to decide. The offer comes two days before quarterly earnings and covers gaming, Windows, Copilot, HR, and finance, while marketing, commercial business, and security are excluded.
The voluntary exits do not mean Microsoft will actually lay off a quarter of its people. 25% is the number of people offered the deal, not the number who will accept it. Microsoft has also been cutting staff for several years, and the timing may have as much to do with AI infrastructure costs as with team performance.
Source: The Seattle Times
OpenAI Launches GPT-6.1 Terra
OpenAI released GPT-6.1 Terra, a faster, cheaper model aimed at everyday computer work — reading invoices, filling in forms, working with technical drawings, or editing video. OpenAI says errors on its internal benchmark fell from 53.6% to 16.7% at low reasoning effort, and standard API pricing is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. For the hardest coding tasks, Sol and Astra remain the better choice.
The model sits in a line-up — Luna < Terra < Sol < Astra — that maps mainly to model size: bigger models are more reliable and recall obscure facts better, but cost more per token. OpenAI says plain “small/medium/large” labels would not age well as models grow, so the astronomy names remain a rough guide rather than a precise description of capability. The launch benchmarks are still worth reading with caution, since vendors naturally highlight the tests where their newest model shines.
Source: OpenAI
Apple Announces iMac With M6
Apple announced a new iMac with the M6 chip on Tuesday, available November 6, with pre-orders starting October 28. It keeps the 24-inch 4.5K Retina display and comes in seven colours, with up to a 12-core CPU and GPU, Neural Accelerators in every GPU core, a new Dual 16-core Neural Engine, and up to 32GB of unified memory. The 10-core model starts at $1,499; the 12-core version starts at $1,699.
The iMac still tops out at a 24-inch panel and 32GB of memory, so it remains an all-in-one for everyday work and local AI rather than a pro workstation. At $1,699, the comparison with a Mac mini and a separate display is inevitable — the iMac’s advantage is the clean single-cable setup, not raw value.
Source: Apple Newsroom
Python’s Experimental JIT Is Being Removed
CPython’s experimental JIT compiler, added in 2024 for Python 3.13, is on its way out. In June the Python steering council paused work on the JIT in the main branch and asked for a standards-track PEP covering maintenance, compatibility, and clear success metrics. The maintainers drafted PEP 836, but withdrew it on October 19, saying they did not have the personal bandwidth to carry it forward. The code will likely be removed from CPython main soon, early in the Python 3.16 development cycle.
The JIT was also meant to offset some of the performance cost of free-threaded Python; without it, that trade-off remains harder to manage. Alternatives such as PyPy and GraalPy exist, but they are not the reference implementation and can struggle with extension compatibility.
Source: LWN.net
Google Launches Gemini 4 Carbon, Starting With Cyber Defenders
On October 28 Google announced Gemini 4 Carbon, its strongest model yet, which the company says ranks first on the updated Artificial Analysis Intelligence Index v5.0 with a score of 59. On security benchmarks it reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for validating exploitability, and 71.0% on CWE-bench for writing patches; on WebDev Arena it gained 89 elo points to reach 1768. Carbon follows Gemini 4 Argon from mid-October and launches at the same introductory price of $2 per million input tokens and $10 per million output tokens, which doubles when the introductory period ends.
The model is not publicly available yet. Google is giving it to a subset of cyber defenders this week and will add more partners through its Fairwind programme, while taking part in the US government’s voluntary pre-release testing. Google also reports a 0.0% success rate for every evaluated indirect prompt-injection attack on the Gray Swan benchmark, where Argon scored 0.7%.
Source: Google
Apple Unveils a Touchscreen MacBook Pro With a PROLED XDR Display
On October 28 Apple introduced a MacBook Pro with a new PROLED XDR display built on a tandem OLED panel: up to 2,000 nits of peak HDR and SDR brightness, a 10,000,000:1 contrast ratio, a 120Hz refresh rate, and coverage of both P3 and Adobe RGB. Touch comes to the Mac for the first time, with the sensor built into the panel itself, and the Dynamic Island arrives on the Mac as the home for Spotlight, Live Activities, and Siri AI. The laptop uses M5 Pro and M5 Max chips, up to 128GB of unified memory and 16TB of storage, and weighs 3 pounds in the 14-inch size and 4 pounds in the 16-inch size; the 14-inch model starts at $2,899 with M5 Pro and $4,499 with M5 Max.
A day earlier, Apple showed the cheaper 14-inch MacBook Pro with the M6 chip: a 12-core CPU (two more cores than M5), a 12-core GPU with Neural Accelerators in every core, and a Dual 16-core Neural Engine. Apple promises up to 30% faster graphics and up to 25% faster AI workloads than M5, 16GB of unified memory as standard (32GB maximum), up to 170GB/s of memory bandwidth, and up to 24 hours of battery life. Pre-orders for both models opened October 28, sales start November 6, and the M6 version starts at $1,999.
Touch remains exclusive to the more expensive PROLED model — the cheaper M6 keeps the older mini-LED display and has no touch, but costs $900 less. macOS translates touches into clicks, so most existing apps work unchanged, and Apple redesigned the hinge so the display stays steady while being touched.
Source: Apple Newsroom, Apple Newsroom
Stack Overflow Falls Below 1,000 New Questions a Month
According to a query in the Stack Exchange Data Explorer, Stack Overflow received just 952 new questions in October 2026, down from 969 in September. A year ago it was 5,263 questions, in October 2024 it was 23,162, and in October 2021 it was still 118,765. The decline has continued in the years since part of the traffic moved to large language models.
Question volume is not the same as traffic, and many common problems already have an answer. The site’s strict duplicate policy also sent new questions to older discussions, and the pace of decline accelerated after 2022: from 105,743 questions in October 2022 to 52,334 a year later, 23,162 in 2024, and 5,263 last year.
Source: Stack Exchange Data Explorer
A Kentucky Humanist Wins an Exemption From Using AI at Work
David Durman, a 45-year-old research administrator at the University of Kentucky, applied this year for a religious accommodation under the federal Title VII law, arguing that using tools such as Copilot, Claude, or ChatGPT would violate his humanist convictions. The American Humanist Association helped with the request, and the university granted it on September 25: he does not have to use AI for his work, but he must still attend meetings and cooperate with colleagues who do. The association describes it as the first such case for a humanist.
Federal courts recognised humanism as a religion back in the 1980s, so the same rights apply as to other faiths; the key questions are whether the belief is sincerely held and whether it imposes an undue burden on the employer. Durman’s case fits a wider shift in public opinion — a Washington Post-ABC-Ipsos poll published on October 21 found that 57% of Americans think AI is bad for society and two-thirds oppose building the data centres needed for it.
Source: Religion News Service
Binance Loses $6.3 Billion in the Largest Crypto Theft on Record
On Thursday, October 29, Binance said an attack had cost it roughly $6.3 billion in crypto assets, which security firms described as the largest crypto theft in history. The exchange’s initial investigation pointed to a compromised DevOps engineer’s laptop that let an attacker reach a service container and submit unauthorised withdrawal transactions; Binance said it found no evidence that any wallet private keys leaked. By Saturday about $3 billion had been frozen, including $2.3 billion in USDT and USDC through Tether and Circle, and Binance said it had replenished 90% of the stolen assets and would absorb the losses through company funds and its SAFU reserve.
On Saturday, October 31, the FBI attributed the attack to North Korea’s TraderTraitor group and published more than 600 Ethereum addresses used to launder the funds. The 18,270 stolen bitcoin, worth about $1.4 billion at the time, were split across 19 addresses — more bitcoin than Tesla holds. The attacker also posted part of the stablecoins as collateral on the Aave lending protocol and borrowed at least 20,000 WETH worth roughly $44 million; when the issuers froze that collateral, the loss landed on the lenders.
Source: Arkham Intelligence, CoinDesk
A Tampered Claude Code Extension Shipped an SSH Backdoor
On October 30 the security firm Phorion documented a backdoor in the official Claude Code for VS Code extension bundle on the Cursor Marketplace, specifically versions 2.1.312 and 2.1.313 for macOS and Windows. The injected code runs as soon as the extension loads, with no action from the user: it appends an attacker-controlled public key tagged svc-telemetry to ~/.ssh/authorized_keys, tries to enable SSH, opens a reverse tunnel to 134.209.82.225, and installs a LaunchAgent named com.apple.sysmond so the tunnel comes back after a reboot. Cursor says the tampered versions were live for about eleven hours on October 29, between 2:27am and 1:35pm Pacific time; the company pulled them, shut the marketplace down as a precaution, and had the attacker’s server taken offline by its cloud provider.
The payload has a bug: without administrator rights systemsetup -getremotelogin prints an error, the script reads that as “SSH is already running” and never enables the service, so on a Mac without Remote Login already on, the tunnel hits a closed port 22. The key, the persistence and the host reconnaissance still land. Anthropic said the incident was not a compromise of its systems and that versions on its official channels are unaffected; the problem arose on the marketplace side, where the extension carried its name.
OpenAI Merges ChatGPT and Codex Usage Limits Into One Allowance
In a help-centre article OpenAI said that from December the Plus, Pro and Business plans will share a single usage allowance across ChatGPT and Codex, as the agentic features of Work move into the main ChatGPT experience. Usage will be measured in tokens: more advanced models and complex tasks draw more from the allowance, while Instant chat stays unlimited and does not count against it. The combined remaining usage and reset times will appear in Settings → Usage, and once the allowance runs out users can wait for a reset, upgrade, or buy credits. In an email about the change, OpenAI said it will test a combined Chat and Work experience with a small group of users over the coming weeks and that Codex will remain a separate experience in the ChatGPT desktop app.
OpenAI says most users should not need to wait for their usage to reset more often than they do today — though “most” also implies that some will. Estimating consumption in advance is hard, because it depends on the model, the length of the context, and how much reasoning and tool use the task triggers; OpenAI points developers to its documentation and tokenizer.
Source: OpenAI Help Center