Amazon and Anthropic Face AI Accountability Questions as Workers Seek a Voice

AI adoption is moving faster than governance and safeguards in several fields. A survey of 220 software practitioners, published in Discover Computing, found 70.9% use AI tools at least weekly, while 61.8% have never received formal training on AI governance or security risks. Researchers Arun Karthik Dharmalingam and T R Kalai Lakshmi developed the AI Governance Deficit Index, which averaged 44.80 out of 100. They cautioned that it summarizes survey responses and is not a validated measure of organizational maturity.

In healthcare, a systematic review of 29 publications from March 2021 to March 2026 found that neurosurgery AI models can report strong accuracy but often lack external testing and reliable calibration. The review covered 23 predictive-model studies, two large-language-model benchmarks and four reviews or methods assessments. Neuro-oncology and cerebrovascular neurosurgery each accounted for six model studies. The authors identify validation and calibration gaps as obstacles to safe clinical use.

Business and technology leaders are also weighing how AI changes operations and products. T-Mobile CEO Srini Gopalan, speaking with Stephanie Mehta, discusses employee engagement, AI and a customer-first mindset as parts of the company’s strategy. Ericsson and Altair Semiconductor report a first-of-its-kind connection using enhanced Reduced Capability (eRedCap) for 5G. AI processing could run on a device or through cloud models, but the companies have not demonstrated AI workloads or published speed, latency or energy measurements; commercial-network and device testing remains ahead.

For software development, Cognition launched macOS support for its AI software engineer, Devin, on September 15, 2026. Each session runs in a fresh macOS virtual machine on Amazon EC2 Mac, with snapshots stored in Amazon S3. About 90% of sessions release the virtual machine when idle and resume from a snapshot in seconds. Devin can build, run and test iOS and macOS apps, and users can view and control them. Nearly 1,000 organizations were using Devin on macOS within six weeks of launch.

AI’s effects on workers and creators remain contentious. Berlin interpreter Sara Campos lost a longtime client after it switched to AI translation; graphic designer Adrian Romero, who had helped introduce AI tools at his agency, lost his job and began moving to insurance work after six months of job hunting. Voice actor Matti Klemm is taking legal action against Netflix, alleging it used his voice to train AI without permission. IAM Union President Brian Bryant said at a November 19, 2026, labor-and-technology panel that AI can be useful when it improves safety or helps people work, but called for protections against worker replacement. He said collective bargaining can give workers a voice, and criticized the absence of unions and workers from a recent White House AI Summit attended by corporations. Bryant also argued that some automation offers no safety or public benefit and can be rejected, citing Waymo.

Concerns about AI safety and accountability are also drawing scrutiny. Safety voices warn that a disaster or rapid capability growth could arrive within months or by 2027, though a prediction of a government overthrow as soon as May is described as an outlier. A report from nonprofit METR, whose researchers spent 11 weeks at Anthropic studying AI development, adds to worries that AI could accelerate its own development and become difficult to control. Anthropic withheld key data, raising questions about how readily the findings can be checked and METR’s independence. Separately, federal contractors could face False Claims Act scrutiny for exaggerating AI capabilities, misreporting milestones or falsely certifying cybersecurity protections. Investigators may compare claims with tests, deployment records and internal warnings; the article advises documenting tests and distinguishing working features from development goals.

Consumer devices are another visible part of the shift. A November 20, 2026, TechCrunch roundup names six AI-oriented home-office products: the $159 Flowtica Scribe Pen, which records and summarizes meetings and supports 15 speakers and 92 languages; the $159 Anker Soundcore Work recorder, with up to eight hours of recording (32 with its case) and 300 free transcription minutes monthly; Logitech MX Keypad; Kindle Scribe Colorsoft; the $249 Insta360 Link 2 Pro webcam, with tracking and automatic framing; and the $599 Autonomous Desk Pro, which tracks sitting and standing and can remind users to move.

Key Takeaways

  • In a 220-person software survey, 70.9% use AI weekly or more, while 61.8% report no formal AI governance or security training; the index average was 44.80/100 and is not a validated maturity measure.
  • A review of 29 neurosurgery AI publications found gaps in external validation and calibration that may hinder safe clinical use.
  • T-Mobile CEO Srini Gopalan identifies employee engagement, AI and customer focus as elements of the company’s strategy.
  • Ericsson and Altair completed an eRedCap 5G connection, but have not published performance measurements or demonstrated AI workloads.
  • Cognition’s Devin runs macOS sessions on Amazon EC2 Mac and stores snapshots in Amazon S3; nearly 1,000 organizations used the macOS product within six weeks of its September 15, 2026 launch.
  • Creative workers report lost work linked to AI; voice actor Matti Klemm is taking legal action against Netflix over alleged unauthorized use of his voice for AI training.
  • IAM Union President Brian Bryant says workers need a voice in workplace AI decisions, including through collective bargaining, and supports AI that improves safety or helps workers.
  • METR’s report on AI development, based on 11 weeks of research at Anthropic, faces questions about withheld data and independent verification; some safety warnings anticipate major risks by 2027.
  • Federal contractors could face False Claims Act scrutiny for overstated AI capabilities, inaccurate development claims or false cybersecurity certifications.
  • Home-office products highlighted include Flowtica Scribe Pen and Anker Soundcore Work at $159 each, Insta360 Link 2 Pro at $249, and Autonomous Desk Pro at $599.

Survey finds AI use outpacing training and safeguards in software teams

A survey of 220 software practitioners, published in Discover Computing, found that 70.9 percent use AI tools at least weekly, while 61.8 percent have never received formal training on AI governance or security risks. Researchers Arun Karthik Dharmalingam and T R Kalai Lakshmi created the AI Governance Deficit Index to measure shortfalls across four workplace areas. The average score was 44.80 out of 100, with higher scores showing greater deficits. The authors stressed that the index is a summary of survey responses, not a validated measure of organizational maturity.

Review finds gaps in testing and calibration of neurosurgery AI models

A systematic review of research published from March 2021 to March 2026 found that neurosurgery AI models often show strong accuracy but lack external testing and reliable calibration. The review followed PRISMA 2020 guidelines and was registered in PROSPERO as CRD420261373632. Researchers examined 29 publications, including 23 predictive model studies, two large language model benchmarks, and four reviews or methods assessments. The 23 model studies covered six areas, with neuro-oncology and cerebrovascular neurosurgery leading at six studies each. The authors said weak validation and calibration remain barriers to using these tools safely in clinical care.

T-Mobile CEO explains the company’s plan for the AI era

T-Mobile CEO Srini Gopalan spoke with Stephanie Mehta about how the company plans to stay ahead as artificial intelligence grows. He discussed employee engagement, AI, and a customer-first mindset as parts of the company’s strategy.

Ericsson and Altair test a lower-power 5G connection for AI wearables

Ericsson and Altair Semiconductor completed a network connection using enhanced Reduced Capability, or eRedCap, which they describe as the first of its kind. The modem provides connectivity, while AI processing can happen on a device or through cloud models. The companies did not show AI workloads or publish measurements for speed, delay, or energy use, and they did not name carriers, device makers, or markets. Commercial deployment is targeted for later years, with testing on commercial networks and devices still needed to show how the trade-offs work in practice.

Creative workers push back as AI takes jobs and raises legal concerns

AI is changing creative work, and some workers say they have lost jobs or clients because of it. Berlin interpreter Sara Campos lost a longtime client after the company switched to AI to cut costs and translate into several languages at once. Graphic designer Adrian Romero helped bring AI tools to his agency, then lost his job and, after six months of job hunting, began moving to an insurance company. Voice actor Matti Klemm is taking legal action against Netflix, claiming it used his voice to train AI without his permission.

AI safety leaders warn that major risks may arrive much sooner

Some influential AI safety voices now warn that an AI disaster or rapid growth in AI capabilities could happen within months or by 2027. One prediction that a government could be overthrown as soon as May is described as an outlier. The article focuses on a report by the nonprofit METR, whose researchers spent 11 weeks at Anthropic studying AI development. Anthropic withheld key data, raising questions about how well the findings can be checked and how independent METR can be. The report adds to concerns that AI systems could speed up their own development and become difficult to control.

How Cognition runs Devin coding agents on Amazon EC2 Mac

Cognition launched macOS support for its AI software engineer, Devin, on September 15, 2026. Each session gets a fresh macOS virtual machine on Amazon EC2 Mac so Devin can build, run, and test iOS and macOS apps. Cognition uses snapshots stored in Amazon S3, and about 90% of sessions give up their Mac virtual machine while idle and resume from a snapshot in seconds. Its tools prepare macOS, install development software, and let users view and control apps, while Devin can use an accessibility tree to identify screen controls. Within six weeks of launch, nearly 1,000 organizations were running Devin on macOS.

IAM Union president calls for worker input on AI at work

On November 19, 2026, IAM Union President Brian Bryant joined labor and technology leaders at a College of Employment and Labor Lawyers panel about AI and the future of work. Bryant said the union supports AI that improves safety or helps people do their jobs, but called for protections against systems that replace workers. He said collective bargaining agreements can give members a voice in how and when employers use AI. Bryant also noted that workers and unions were not invited to a recent White House AI Summit attended by corporations. He said some forms of automation and AI bring no safety or public benefit, adding that it can be acceptable to reject them, as in the case of Waymo.

False AI claims in federal contracts could draw enforcement action

Federal contractors may face False Claims Act scrutiny if they exaggerate AI capabilities or falsely certify AI-related cybersecurity protections. Examples include claiming drones can track targets on their own when people still intervene, or overstating a medical device’s diagnostic abilities. Contractors could also misreport AI development milestones, bill for work they did not perform, or claim that cybersecurity tools are active when they remain in testing. Investigators may compare claims with test results, deployment records, and internal warnings. The article advises contractors to review AI claims, document testing, and clearly distinguish working features from development goals.

Six AI gadgets aim to upgrade home offices

A November 20, 2026, TechCrunch article highlights six AI-powered devices for home offices. The Flowtica Scribe Pen costs $159 and can record meetings, transcribe them, summarize key points, and create to-do lists; it supports 15 speakers and 92 languages. Anker Soundcore Work costs $159, records for up to eight hours or 32 hours with its case, and offers 300 free transcription minutes each month. Other products include Logitech MX Keypad, Kindle Scribe Colorsoft, the $249 Insta360 Link 2 Pro webcam, and the $599 Autonomous Desk Pro. The webcam uses tracking and automatic framing, while the desk tracks sitting and standing and can remind users to move.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI governance AI adoption AI security training AI governance deficit healthcare AI neurosurgery AI clinical validation AI calibration 5G eRedCap AI software engineering Cognition Devin Amazon EC2 Mac AI and employment labor rights collective bargaining AI voice training consent AI safety AI accountability False Claims Act AI consumer devices

Comments

Loading...