England’s Office for Students (OfS) warns that AI tools, including ChatGPT, can produce strong answers to some university assessments in minutes, putting the reliability and credibility of those assessments at risk. Research with Advance HE involved 244 higher-education staff and 91 students from at least 127 universities; respondents described similar-looking submissions and difficulty detecting AI use. The regulator urges universities to ensure assessments test students’ knowledge and skills, and says institutions could face sanctions if they fail to do so.
Universities are responding with more supervised and in-person work, including written and practical exams, presentations, oral exams known as vivas, and exercises asking students to evaluate AI output or explain their process. The National Union of Students says institutions should also move beyond learning methods it considers outdated. A separate analysis of AI policies at 39 universities finds wide variation: rules often sit in misconduct policies rather than assessment guidance, and only one university had published specific AI guidance. Its author argues that assessment and curriculum design need to change, not just rules on permitted AI use.
Girish Konda says AI subagents need safeguards drawn from distributed-systems engineering. Multiple agents repeating the same guess do not amount to independent confirmation, and systems must account for slow workers being mistaken for failed ones or crashes obscuring exit codes. Konda says SAFE cannot make a claim true or guarantee exactly-once delivery; he recommends refusing to guess and using a single orchestrator. Simple drafting assistants with human oversight may require fewer protections.
AI also raises questions about trust in finance: deepfakes and voice cloning can undermine identity checks and confidence in audio or video evidence, while decisions made by AI agents may be difficult to understand or audit. The article calls for policy to adapt to these risks. Separately, ByteDance is reportedly developing an AI assistant to compete with US labs. The report says it may serve a role similar to Muse, does not clarify whether it will be paired with hardware, and notes that ByteDance’s Seed models use smaller datasets than frontier competitors. It also cites a survey finding that 79% of consumers consider plain-language explanations of an agent’s choices very important or essential.
In research, Exeter protein-motion scientist Dr JJ Phillips is among 20 new UKRI-funded Turing AI Fellows. The programme has more than £39 million in government funding: 17 researchers receive Turing AI Pioneer Interdisciplinary Fellowships and three receive Turing AI Global Fellowships. Phillips will use experiments tracking protein movement to help AI design tools create biosensors and potentially more advanced molecular machines.
A.I.Matics says its AIM-T1 inspection platform has received CES 2027 Innovation Awards Honoree recognition in both Artificial Intelligence and Construction & Industrial Tech. Designed for electronics manufacturing, it inspects products including camera modules, semiconductor packages and printed circuit board assemblies. Its modular imaging hardware, software-defined inspections, world model technology and in-context learning use acceptable product samples to establish criteria and detect differences. The company plans to show it at CES in Las Vegas, Jan. 6–9.
Adobe will demonstrate Project Live Set at its Adobe MAX 2026 Sneaks event on Nov. 11. The early-stage video-editing research project lets users change a scene’s subject, setting, characters or camera direction while it is being generated, with results appearing in real time. Adobe’s demo uses prompts to transform a black cat into other characters and animals. The company has not disclosed the models behind it and says the preview does not promise a future product feature.
In Korea, Lotte Innovate and Thales have agreed to connect Thales-owned Imperva web security technology with Lotte Innovate’s security operations platform. The memorandum, signed at Lotte Innovate’s Seoul headquarters on Oct. 28 and announced Nov. 11, covers broader application security across Lotte affiliates, improved attack detection and response, integrated security services and joint marketing. Lotte Innovate says it will launch an AI premium security monitoring service this month, pursue new customers with Thales, and build on more than 1,000 completed security projects as it expands to external businesses.
Key Takeaways
- England’s OfS says AI can complete some university assessments to a high standard in minutes and warns institutions could face sanctions if assessments do not properly test student knowledge and skills.
- Universities are adding supervised exams, presentations, vivas and tasks that ask students to assess AI output; an analysis of 39 universities finds policies are often vague, and only one had specific AI guidance.
- Girish Konda recommends distributed-systems safeguards for AI subagents, including a single orchestrator and refusing to guess; repeated agent guesses are not independent confirmation.
- Deepfakes, voice cloning and opaque AI-agent decisions pose risks to trust in finance, prompting calls for policy to adapt.
- ByteDance is reportedly developing an AI assistant to compete with US labs; the report says its Seed models use smaller datasets than frontier competitors and cites 79% of consumers valuing plain-language explanations of agent decisions.
- Exeter researcher Dr JJ Phillips is one of 20 UKRI-funded Turing AI Fellows in a programme backed by more than £39 million; his work applies protein-motion experiments to AI-designed biosensors and potentially molecular machines.
- A.I.Matics’ AIM-T1 has CES 2027 Innovation Awards Honoree recognition in Artificial Intelligence and Construction & Industrial Tech for electronics inspection; the company plans to exhibit Jan. 6–9 in Las Vegas.
- Adobe’s Project Live Set preview at Adobe MAX 2026 Sneaks shows real-time scene changes during AI video generation, but Adobe says it is an early research project, not a promised product feature.
- Lotte Innovate and Thales plan to integrate Imperva technology with Lotte Innovate’s security platform; Lotte Innovate says it will launch an AI premium security monitoring service this month.
Regulator urges universities to rethink assessments as AI use spreads
England’s Office for Students (OfS) says universities should use more traditional exams and in-person presentations because AI makes some coursework hard to assess reliably. It warns that institutions could face sanctions if their assessments do not properly test students’ knowledge and skills. The OfS says many universities are already adding in-person written exams, oral exams called vivas, and tasks that ask students to review AI output or explain how they did their work. A survey found students using AI to finish assessments in minutes, while staff reported similar-looking submissions and concerns that some assessments no longer work well.
OfS warns students can complete university assessments with AI in minutes
England’s Office for Students (OfS) says many university assessments are unsupervised and can be completed to a high standard with AI, sometimes in minutes. Its research with Advance HE drew on 244 higher education staff and 91 students from at least 127 universities. The OfS says AI use can be hard to detect and threatens the credibility of assessments, while closed-book and practical exams are among the most effective options. Universities UK says many institutions have changed their assessments, including by adding in-person exams, presentations, vivas and controlled uses of AI. The National Union of Students argues that universities should also move beyond traditional learning methods that it says are outdated.
Girish Konda says AI subagents need distributed systems safeguards
Girish Konda argues that AI subagents should be treated like parts of a distributed computer system. He says several agents repeating the same guess do not provide independent confirmation, and cites distributed-systems ideas including Lamport’s happened-before rule and weighted voting. Konda describes risks such as timeouts that mistake a slow worker for a failed one and crashes that hide an exit code. He says SAFE does not make facts true or guarantee exactly-once delivery, and recommends that it refuse to guess and use one orchestrator. Simple draft helpers supervised by a person may need fewer safeguards.
Study finds university AI rules are often vague and limited
A November 10, 2026 article by Jason M. Lodge says universities have struggled to adapt to generative AI since ChatGPT arrived in late 2022. Large classes and limited contact between students and teachers can make it harder for universities to know students and confirm they learn what their degrees require. An analysis of 39 universities found wide differences in AI policies, with rules often placed in misconduct policies rather than assessment design guidance. Only one university had published specific guidance on AI, and general guidance was often vague. The article argues that universities need changes to assessment and curriculum, not only rules about when students can use AI.
AI threats to trust could reshape the future of finance
The article argues that trust is central to modern finance and that artificial intelligence creates new risks for the financial system. Deepfakes and voice cloning can make it harder to confirm someone’s identity or trust audio and visual evidence. Another risk is that people may not understand or be able to audit the decisions made by AI agents. The article says policy must adapt to protect trust as these technologies emerge.
Exeter researcher joins 20 UKRI Turing AI Fellows
Dr JJ Phillips, an Exeter academic who studies how proteins move, is one of 20 new UKRI-funded Turing AI Fellows. The fellows will use artificial intelligence to address challenges such as improving disease diagnosis and developing new treatments. The program has more than £39 million in government funding, with 17 researchers supported by the Turing AI Pioneer Interdisciplinary Fellowships and three by the Turing AI Global Fellowships. Phillips will use experiments that track protein motion to help AI design tools create biosensors and, potentially, more advanced molecular machines.
A.I.Matics AIM-T1 earns two CES 2027 Innovation Awards honors
A.I.Matics says its AIM-T1 AI vision inspection platform has been named a CES 2027 Innovation Awards Honoree in Artificial Intelligence and Construction & Industrial Tech. The system helps electronics manufacturers inspect products such as camera modules, semiconductor packages and printed circuit board assemblies as designs and production needs change. It combines modular imaging hardware, software-defined inspections, world model technology and in-context learning, using acceptable product samples to set inspection criteria and spot differences. A.I.Matics plans to showcase the platform at CES 2027, which will run Jan. 6 to 9 in Las Vegas.
Adobe previews real-time AI video editing with Project Live Set
Adobe will preview Project Live Set during its Adobe MAX 2026 Sneaks event on Nov. 11. The early research project lets video editors change subjects, environments, characters or camera direction while a scene is being generated, with results shown in real time. Adobe says the feature aims to reduce waiting and help creators shape ideas as they work. The demo uses prompts to transform a video of a black cat into different characters and animals. Adobe has not shared details about the models behind the project and says the preview is not a promise of a future product feature.
ByteDance reportedly develops AI assistant to rival US labs
ByteDance is developing a new AI agent that could help the Chinese company compete with AI labs in the United States, according to the article. The report says the assistant will serve a function similar to Muse, though it does not clarify whether ByteDance plans to pair it with hardware. It also says ByteDance’s Seed AI models are trained on smaller datasets than those of frontier competitors, and that Zhang is focused on data. The article notes that consumers want AI agents to explain their choices, with 79% saying it is very important or essential for an agent to explain in plain language why it chose an offer or action.
Lotte Innovate and Thales partner to expand AI security in Korea
Lotte Innovate signed a memorandum of understanding with Thales at its Seoul headquarters on Oct. 28, the company announced on Nov. 11. The agreement connects Imperva web security technology, owned by Thales, with Lotte Innovate’s security operations platform. The companies plan to expand application security across Lotte affiliates, improve attack detection and response, package integrated security services and pursue joint marketing. Lotte Innovate will launch an AI premium security monitoring service this month and work with Thales to attract new customers. The company says it has completed more than 1,000 security projects and plans to expand its services to outside businesses.
Sources
- AI has made it ‘close to impossible’ to assess university coursework, regulator says
- Students passing university assessments in minutes using AI, regulator warns
- Treat subagents as distributed systems
- AI use is now universal in higher education. Why are universities failing to adapt?
- Trust beyond face-to-face contact: AI and the future of finance
- Exeter academic named among 20 new UKRI Turing AI Fellows to strengthen UK AI capability
- A.I.Matics' AIM-T1 Named CES Innovation Awards® 2027 Honoree in Two Categories
- Adobe Project Live Set Real-Time AI Video Editing Feature
- ByteDance Builds Assistant to Take on US AI Labs
- Lotte Innovate signs MOU with Thales to expand AI security services in Korea
Comments
Please log in to post a comment.