Cambridge Dictionary names “human” its 2026 word of the year, reflecting concern that AI is blurring distinctions between people and machines—and that people may lose something by outsourcing their thinking. Lexicographer Colin McIntosh says people across fields are reconsidering their reliance on AI. The word, derived from Latin “humanus” and in use since the 1450s, describes people’s qualities, faults and feelings. Other contenders included “agent” and “analogue.” Microsoft co-founder Bill Gates says AI’s success should be measured by how much it improves human lives; other recent AI-linked dictionary picks include “hallucinate,” “slop” and “vibe coding.”
Schools are testing practical ways to use AI while keeping teachers and students in control. Arcadia Middle School in Wisconsin is piloting a UW-Madison tool that records and transcribes group discussions so teachers can search for ideas, check participation and see whether students stay on task. It can also translate conversations, which matters in a district where 75% of students are English-language learners. In Door County, teachers encourage tools such as Google Classroom to improve work while requiring students to do their own learning. Educators use monitoring and handwritten assignments to curb misuse, but warn that unreliable AI detectors can damage trust; districts are still developing formal rules.
Amazon Web Services outlines ways to reduce the cost and delay of AI agents handling large data payloads. In tests using a 23 MB JSON payload for a Flightscape production agent built with Amazon Bedrock AgentCore, combined strategies cut token use by up to 71% and average latency by up to 80%, while maintaining reasoning across extended sessions. The advice addresses a common issue: large tool outputs can repeatedly fill a model’s context window.
Security and oversight remain pressing concerns as organizations adopt AI. At a Seoul briefing, F5 reported that 63% of surveyed Asia Pacific executives cite unpredictable token costs as a barrier, 53% prohibit sensitive data from leaving their countries, and half say their security tools are not ready for AI. The survey also found 35% worry about leaks from autonomous agents; 93% of companies use hybrid multicloud systems spanning an average of 19 locations, while 78% use an average of seven AI models. F5 urged firms to secure systems across locations and avoid reliance on a single model, noting attacks can exploit vulnerabilities an average of seven days after discovery.
OpenAI has made agreements with Apollo Research, METR, Transluce and Goodfire to independently assess its safety work. The groups will receive access to selected internal models and resources, potentially including source code, training data, monitoring systems and technical documents. Their work will examine issues including whether models act against developers’ intentions, alignment and monitoring, training-time behavior, and post-training evaluations. OpenAI describes the agreements as the first stage of a broader plan and says they support commitments under the White House Accord on Super Intelligence. Separately, researchers introduced GUARD, a framework that lets agents revise prompts, skills, tool rules and memory without changing the underlying model. Tests report improved performance without catastrophic forgetting or memory pollution, alongside substantially lower deployment token use.
Legal and operational risks are also prompting calls for stronger rules and safeguards. Congress has not advanced any of five identified measures on AI and copyright beyond committee referral, while more than 150 U.S. copyright lawsuits involving AI companies have been identified. Courts disagree on whether copyrighted works may be used for training without permission; a September Third Circuit ruling rejected a fair-use defense in a ROSS Intelligence case but distinguished generative AI cases. In legal practice, iManage’s Jan Van Hoecke argues firms should contain hallucination risks rather than wait for better models. A public database had recorded more than 2,100 court decisions involving fabricated or misrepresented legal authority by October 2026. He recommends checking primary sources before work is filed or shared, tracking errors and keeping audit records, especially when agents can take actions.
Police departments face a similar gap between adoption and preparation. A National Policing Institute report, discussed by its president Jim Burch on the Policing Matters podcast, found that 83% of participating agencies had deployed at least one AI tool, while 44% had offered no AI-specific training. Burch calls for a national strategy led by policing professionals, with agencies starting small, testing tools, building community support and planning their first 90 days. Another article questions whether AI superintelligence is technically achievable, describing arguments that it is speculative or may not be computable, alongside concerns about self-improving systems and potential harms. It calls for independent engineers and scientists to join public debate, rather than leaving it to a small group of companies, including OpenAI and Anthropic, that pursue superintelligence.
Key Takeaways
- Cambridge Dictionary’s 2026 word of the year is “human,” reflecting concerns about AI, reliance on automated thinking and what distinguishes people from machines; Bill Gates says AI should be judged by its impact on human lives.
- Arcadia Middle School is testing a UW-Madison discussion tool that records, transcribes and translates classroom conversations; 75% of district students are English-language learners.
- Door County educators encourage responsible use of tools such as Google Classroom but caution that AI detectors can wrongly accuse students and undermine trust.
- AWS reports that combined strategies cut token use by up to 71% and average latency by up to 80% in tests involving a 23 MB JSON payload for an Amazon Bedrock AgentCore agent.
- F5’s Asia Pacific survey found 63% cite unpredictable token costs, 53% restrict sensitive data from leaving their countries, and 35% worry about autonomous-agent data leaks; 78% use an average of seven AI models.
- OpenAI has given Apollo Research, METR, Transluce and Goodfire access to selected internal resources for independent assessments of its AI safety work.
- GUARD aims to let AI agents improve prompts, skills, rules and memory without changing the underlying model; researchers report gains without catastrophic forgetting or memory pollution.
- More than 150 U.S. copyright lawsuits involving AI companies have been identified, but none of five congressional measures has moved beyond committee referral.
- By October 2026, a public database had recorded more than 2,100 court decisions involving fabricated or misrepresented legal authority; iManage’s Jan Van Hoecke urges primary-source checks and audit records.
- A policing report found 83% of participating agencies had deployed at least one AI tool, while 44% had no AI-specific training; a separate article questions claims about superintelligence and calls for broader public debate beyond companies such as OpenAI and Anthropic.
Cambridge Dictionary names human its word of the year amid AI concerns
Cambridge Dictionary chose “human” as its word of the year, reflecting debate about how artificial intelligence is blurring the lines between people and machines. The choice also reflects fears about what people may lose when they rely on AI to think for them. The article describes the selection as a response to concerns about AI and what it means to be human.
Cambridge Dictionary names human its word of the year amid AI concerns
Cambridge Dictionary chose “human” as its word of the year amid concerns about artificial intelligence. The choice reflects debate about how AI has blurred the lines between machines and people. The dictionary’s selection also points to fears about what humans may lose when they outsource their thinking to AI.
Cambridge Dictionary names human its 2026 word of the year
Cambridge Dictionary chose “human” as its word of the year for 2026, reflecting concern about artificial intelligence and what it means to be human. Lexicographer Colin McIntosh said people are reconsidering how much they rely on AI, from mathematicians to teachers of young children. The word comes from the Latin “humanus,” has been used since the 1450s, and means having the qualities, faults and feelings people have, unlike gods, animals or machines. Other dictionaries have chosen AI related terms, including “AI” in 2023, “hallucinate,” “slop” and “vibe coding.” Microsoft co-founder Bill Gates said AI’s success should be measured by how much it improves human lives. The other 2026 contenders included “agent” and “analogue.”
Cambridge Dictionary picks human amid growing debate over AI
Cambridge Dictionary named “human” its word of the year for 2026, as concerns about artificial intelligence prompt debate over the differences between people and machines. Lexicographer Colin McIntosh said fears about what people may lose by outsourcing their thinking have led them to reconsider how much they rely on AI. Cambridge defines “human,” from the Latin “humanus” and in use since the 1450s, as having the qualities, faults and feelings people have, unlike gods, animals or machines. Other recent dictionary choices linked to AI include “AI,” “hallucinate,” “slop” and “vibe coding.” Bill Gates said AI’s success should be measured by how much it improves human lives. Other contenders for 2026 included “agent” and “analogue.”
Cambridge Dictionary chooses human as 2026 word of the year
Cambridge Dictionary selected “human” as its word of the year for 2026, reflecting concerns about artificial intelligence and its effect on how people think about being human. Lexicographer Colin McIntosh said fears about what people may lose by outsourcing their thinking have led them to reconsider reliance on AI. The word comes from the Latin “humanus,” has been used since the 1450s, and refers to people’s qualities, faults and feelings, as opposed to those of gods, animals or machines. Recent dictionary choices tied to AI include “AI,” “hallucinate,” “slop” and “vibe coding.” Bill Gates said AI’s success should be measured by how much it improves human lives, while other 2026 contenders included “agent” and “analogue.”
Cambridge Dictionary names “human” its 2026 word of the year
Cambridge Dictionary chose “human” as its word of the year on November 17, 2026, amid concerns about artificial intelligence. The choice reflects debate over how AI is changing ideas about what makes people different from machines. The dictionary’s lexicographer Colin McIntosh said AI is changing how people talk about being human.
Cambridge Dictionary names “human” its word of the year
Cambridge Dictionary selected “human” as its word of the year, reflecting concerns about artificial intelligence. Lexicographer Colin McIntosh said AI is changing how people talk about being human. Cambridge defines the adjective, which comes from the Latin “humanus” and has been in use since the 1450s, as having the qualities, faults and feelings of people, rather than gods, animals or machines. Bill Gates said AI is making people think harder about what it means to be human and that AI should be judged by how much it improves human lives. AI has also shaped other word of the year choices, including Collins Dictionary’s “AI” in 2023 and Merriam-Webster’s “slop” last year for low-quality AI-produced content.
Wisconsin school tests UW-Madison AI tool to support classroom teaching
Arcadia Middle School in Wisconsin is testing an AI tool developed by the University of Wisconsin-Madison to help teachers follow student discussions during group activities. Devices record and transcribe conversations in real time, letting teachers search for key ideas and see whether students stay on task and take part. Assistant professor Shamya Karumbaiah created the tool and says classroom AI should support teachers and students under teacher oversight. The technology can also translate discussions in different languages, a feature the district values because 75% of its students are English-language learners. Teachers say the tool should support, not replace, their work and must be used responsibly and transparently.
Door County schools weigh AI’s classroom benefits and risks
Door County schools are working out how students and teachers should use artificial intelligence. Sevastopol teacher David Phillips encourages students to use tools such as those in Google Classroom to improve work, while still doing their own research and learning. Districts use device monitoring, screen controls and assignments such as handwritten work to limit misuse, but educators warn that AI detection can wrongly accuse students and harm trust. School leaders say conversations between teachers and students help them judge whether work reflects a student’s learning. Districts are at different stages of setting formal AI rules, with Sevastopol considering policies recommended by consulting firm NEOLA.
Amazon Bedrock strategies cut AI agent token use and delays
Amazon Web Services describes ways to reduce token costs and latency for agents that handle large JSON data with Amazon Bedrock AgentCore. Large payloads can fill an AI model’s context window each time a tool is used, raising costs and slowing responses, especially over long conversations. The post shares three strategies used to build a production agent for Flightscape that handles JSON payloads larger than 20 MB. In benchmark tests on a representative 23 MB payload, the combined changes reduced token use by up to 71% and average latency by up to 80%, while maintaining reasoning ability across extended sessions.
F5 urges firms to prepare for faster AI powered cyberattacks
At a November 18 briefing in Seoul, F5 described three security challenges for companies: protecting data as sovereign AI grows, safely using multiple AI models, and responding to faster AI driven cyberattacks. An F5 survey found that 63% of Asia Pacific executives cited unpredictable token costs as a barrier to wider AI use, while 53% said they do not allow sensitive data to leave their countries. The survey also found that half of companies felt their security tools were not ready for AI, and 35% worried about data leaks from autonomous AI agents. F5 said 93% of companies use hybrid multicloud systems across an average of 19 locations, and 78% use an average of seven AI models. The company urged firms to secure AI systems across locations and avoid depending on a single model, while noting that attacks can exploit vulnerabilities an average of seven days after discovery.
OpenAI gives four independent groups access to test AI safety
OpenAI has signed agreements with Apollo Research, METR, Transluce, and Goodfire to independently assess its AI safety work. The groups will receive access to selected internal models and resources, which may include source code, training data, monitoring systems, and technical documents. Apollo Research will examine whether models might act against their developers’ intentions and whether safeguards would detect this. Goodfire will study models during training, METR will assess alignment and monitoring, and Transluce will review OpenAI’s post training alignment evaluations. OpenAI said the agreements are the first stage of a broader plan and support commitments under the White House Accord on Super Intelligence.
Congress weighs new rules for AI training on copyrighted work
Federal courts have reached different conclusions on whether AI companies can use copyrighted works to train models without permission, leaving Congress to decide whether to create new protections or let courts interpret existing law. More than 150 copyright lawsuits against AI companies have been identified in the United States. A September ruling by the U.S. Court of Appeals for the Third Circuit rejected a fair use defense in a case involving ROSS Intelligence, but the court distinguished its decision from cases about generative AI. Congress is also weighing questions about copyright ownership, which remains tied to human authorship. None of the five congressional measures identified by the Congressional Research Service has moved beyond committee referral, so the legality of AI training will continue to depend on individual cases unless Congress or higher courts set broader rules.
GUARD helps AI agents improve without forgetting earlier skills
Researchers introduced GUARD, a framework that helps autonomous AI agents improve their prompts, skills, tool rules, and memory without changing the underlying model. It uses self checks before tasks and directional regret to identify newly broken behaviors while keeping mastered rules. GUARD compares candidate changes and accepts an edit only when its gains outweigh its regressions, while limiting specialized fixes to avoid overuse. Tests across memory, skill development, ongoing task streams, and domain reasoning showed improved performance without catastrophic forgetting or memory pollution. The approach also substantially reduced token use at deployment, with one frozen harness running in a single pass during testing.
Firms can contain the damage caused by AI hallucinations
Jan Van Hoecke, iManage’s vice president of product management for AI services, says firms should treat AI hallucinations as a risk to contain rather than a problem that a future model release will solve. By October 2026, Damien Charlotin’s public database had recorded more than 2,100 court decisions involving fabricated or misrepresented legal authority. The size of the harm depends on who sees an error, whether it can be corrected, how widely it spreads, how easy it is to detect, and who is accountable. Van Hoecke recommends checks before AI generated work is filed, sent to clients, or saved in shared systems, including checking primary sources rather than only confirming that a case exists. He also advises firms to track where errors are caught and keep audit records, especially when AI agents can take actions such as sending emails or filing documents.
Police agencies need an AI strategy, not just new tools
Artificial intelligence is already built into tools used by many American police agencies, but buying technology does not prepare departments to manage it. In a Policing Matters podcast, host Jim Dudley speaks with Jim Burch, president of the National Policing Institute, about its report on the AI adoption strategy gap. The report draws on an April 2026 roundtable in San Francisco and found that 83% of participating agencies had deployed at least one AI tool, while 44% had offered no AI-specific training. Burch calls for a national strategy led by policing professionals, and discusses how agencies can start small, test tools, build community support, and plan their first 90 days.
Why the push for AI superintelligence may be built on weak claims
The article questions whether AI superintelligence is possible from an engineering perspective, even as some researchers and companies warn that it could threaten humanity. It describes concerns about systems improving themselves, along with arguments that superintelligence is speculative, may not be computable, and could be aligned with human interests through the right design. The author says that pursuing superintelligence may still cause harm, and points to OpenAI and Anthropic as companies pursuing it. The article calls for independent engineers and scientists to take part in public debate rather than letting a small group of AI companies shape the discussion.
Sources
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Cambridge Dictionary unveils its word of the year
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Cambridge Dictionary chooses 'human' as word of the year in response to AI fears
- Arcadia classrooms test UW-Madison AI tool to support teaching
- The Good, the Bad and the AI
- Reducing token costs and latency in Amazon Bedrock AgentCore JSON workloads
- F5 urges multimodel AI and automated security as cyberattacks accelerate
- OpenAI pens AI safety agreement with four firms
- Congress Faces Decisions on AI Training and Copyright Law
- GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance
- Limiting the blast radius of AI hallucinations
- Why police agencies need an AI adoption strategy, not just AI tools
- The AI Superintelligence Debate Has Serious Flaws
Comments
Please log in to post a comment.