AI agents are rapidly moving into real enterprise workflows, handling tasks like querying databases and calling APIs with minimal human oversight. But security practices haven't kept pace. Traditional identity models don't fit how agents operate, and experts say governance around agentic AI still has significant gaps to close.
The federal government released voluntary AI incident response guidance through NIST, CISA, and the NSA on October 16, but security leaders say it focuses too heavily on lab environments. In practice, revoking an agent's access often doesn't stop queued work or related systems, meaning containment strategies need to be far more comprehensive.
Cisco is taking steps to address some of these challenges. The company launched Cisco Model Foundry, a platform for running custom AI models with customer data, complete with API keys, access controls, and model visibility. Cisco is also adding quantum-safe features to its new CW9176ISEC Wi-Fi 7 access point and integrating Model Context Protocol servers to help AI agents access network data through standard interfaces.
New research from METR revealed notable gaps in OpenAI's internal agent monitoring systems. Cyber experts discussed these findings at a recent conference, pointing out that the safeguards were not equipped to handle certain attack types. The results highlight the need for stronger safety measures as AI systems are deployed in critical infrastructure.
On the consumer side, a YouGov survey found that 93% of British adults have at least one concern about AI, with privacy and security topping the list at 65%. Despite those worries, 51% use AI tools at least weekly, and half of those with concerns still engage with AI regularly. The data paints a picture of widespread unease paired with deep adoption.
Google.org is putting $1 million toward a free AI filmmaking program in five African countries. The 12-week course, called From Script to Screen with AI, launches January 25, 2027, and is open to adults in Ghana, Kenya, Nigeria, Sierra Leone, and South Africa. Selected teams can receive funding to produce films between July and October 2027.
The AI security market is consolidating quickly. Analyst Jonathan Care noted that categories like model scanning and AI bills of materials are becoming features inside larger platforms. Cisco acquired Astrix, 1Password bought Apono, and Check Point moved on Cyata as bigger players folded agent security into their offerings. Israeli startups raised at least $2.425 billion in Q3, accounting for nearly half of global cyber venture investment, while 62 seed and Series A rounds across the sector raised $1.16 billion combined.
AI is also surfacing in unexpected places. Dr. Francisco Lara-GarcĂa raised concerns about AI-powered border surveillance on Univision, questioning how continuous monitoring affects border communities and what safeguards govern data collection. In Maryland, school board candidates are voicing concerns about a proposal to use AI for grading and lesson planning, worried it could widen achievement gaps. Meanwhile, an AI-generated endorsement video in Fort Bend County, Texas, has sparked controversy, with Democratic opponent Dexter McCoy calling for a criminal inquiry.
Marissa Mayer's recent AI push drew heavy criticism, reflecting a broader sentiment that people crave genuine human connection. The backlash wasn't about the technology itself but about the feeling that authentic human experiences are being sidelined by automation. As low-quality AI content becomes commonplace, businesses that prioritize heartfelt, people-driven work are finding they stand out.
Key Takeaways
- AI agents are handling real enterprise tasks, but security governance and identity models haven't kept up with their dynamic nature.
- NIST, CISA, and NSA released 10 voluntary AI incident response practices on Oct. 16, but leaders say the guidance is too focused on lab environments.
- Cisco launched Model Foundry for custom AI model deployment and is adding quantum-safe features to its CW9176ISEC Wi-Fi 7 access point.
- METR research found significant safeguard gaps in OpenAI's internal agent monitoring systems, raising concerns about AI safety in critical infrastructure.
- A YouGov survey shows 93% of British adults have AI concerns, yet 51% use AI tools at least weekly.
- Google.org is funding a $1 million free AI filmmaking course in five African countries launching January 2027.
- Israeli startups raised at least $2.425 billion in Q3, nearly half of global cyber venture investment, while 62 early-stage rounds raised $1.16 billion combined.
- AI security categories like model scanning and AI bills of materials are consolidating into larger platforms through acquisitions by Cisco, 1Password, and Check Point.
- AI is sparking debate in border surveillance, Maryland classrooms, and local politics, with concerns about privacy, equity, and authenticity.
- Marissa Mayer faced backlash over AI initiatives, signaling that audiences increasingly value genuine human connection over automated content.
AI agents enter enterprise workflows as security rules lag behind
AI agents now handle real business tasks like querying databases and calling APIs with little human oversight. Security models based on Zero Trust and SASE principles are being tested to manage agent access. Experts stress the need for verified identities, least privilege permissions, and continuous monitoring. Traditional identity categories do not fit agents because they act in dynamic, unplanned ways. Governance around agentic AI is still catching up to the risks.
Federal AI security guidance leaves enterprise threats unaddressed
NIST, CISA and the NSA released 10 voluntary incident response practices for AI systems on Oct. 16. Security leaders say the guidance focuses too much on lab environments and frontier models. In real enterprises, revoking an agent's access often does not stop queued work or related systems. Experts recommend distinct agent identities, short-lived credentials, and continuous investigation. Containment must cover work already handed off to other agents.
Cisco launches AI tools and quantum-safe Wi-Fi to boost security
Cisco introduced Cisco Model Foundry, a platform for running custom AI models with customer data. The service offers API keys, access controls, and visibility across models. Cisco is also integrating ThousandEyes monitoring into Wi-Fi 6E and 7 access points. A new Wi-Fi 7 access point, the CW9176ISEC, will include quantum-safe features aligned to CNSA 2.0. Cisco is adding Model Context Protocol servers to help AI agents access network data through standard interfaces.
Britons worry about AI but half still use it weekly
A YouGov survey found that 93% of British adults have at least one concern about AI. Privacy and security is the top worry at 65%, followed by overreliance at 61%. Still, 51% of adults use AI tools at least once a week. Among those with concerns, 50% still use AI weekly. Only 7% said they have no concerns at all.
Akuna and Google launch free AI filmmaking course in Africa
Akuna Learning and Google opened applications for a free AI filmmaking program in five African countries. The 12-week online course, called From Script to Screen with AI, starts Jan. 25, 2027. Google.org is funding the program with US$1 million. Applicants must be 18 or older and live in Ghana, Kenya, Nigeria, Sierra Leone, or South Africa. Graduates can pitch film ideas, and selected teams receive funding to make films between July and October 2027.
METR research finds gaps in OpenAI's AI agent safeguards
New research from METR found that OpenAI's internal agent monitoring systems had significant safeguard gaps. Cyber experts discussed the findings at a recent conference and stressed the need for stronger AI safety measures. The systems were not equipped to handle certain types of attacks. Experts are calling for more investment in AI safety research and stricter regulations as AI is used in critical infrastructure.
Dr. Francisco Lara-GarcĂa discusses AI border surveillance on Univision
Dr. Francisco Lara-GarcĂa, an assistant professor of computer science, was interviewed on Univision about AI use in border surveillance. He examined how continuous monitoring affects border communities, especially Latino residents. He questioned what safeguards govern surveillance data collection and use. He also raised concerns about AI misinterpreting everyday activities as suspicious behavior and how increased surveillance can erode trust in authorities.
Fort Bend County Judge race heats up over AI endorsement video
An AI-generated endorsement video supporting Republican interim Fort Bend County Judge Daniel Wong has caused controversy. The video shows Black community members endorsing Wong with claims he helped them during difficult times. Wong's campaign says a Democratic supporter created the video with claimed consent from those depicted. Democratic opponent County Commissioner Dexter McCoy condemned the video as exploitation and is seeking a criminal inquiry.
AI security market categories consolidating into bigger platforms
Analyst Jonathan Care discussed AI security market trends in an interview with Help Net Security. He noted that categories like model scanning and AI bills of materials will become features inside larger platforms. He highlighted that Cisco bought Astrix, 1Password bought Apono, and Check Point moved on Cyata to fold agent security into bigger platforms. Q3 funding saw 62 seed and Series A rounds raising $1.16 billion combined. Israeli startups raised at least $2.425 billion in Q3, representing close to half of global cyber venture investment.
Maryland school board candidates express concerns about AI in education
Candidates running for seats on the Maryland State Board of Education are voicing concerns about AI in classrooms. The board is set to vote on a proposal to use AI to help teachers with grading and lesson planning. Some candidates worry AI could reduce the human touch in education and widen achievement gaps between students of different backgrounds. Experts argue AI could revolutionize education by providing personalized learning and freeing teachers to focus on teaching.
Marissa Mayer backlash shows people want real human connection
Marissa Mayer faced strong criticism for her recent AI initiatives, proving that the market is tired of over-automation. The backlash was not about the technology itself but about the timing and the feeling that human life is being forced into efficiency. People are overwhelmed by low-quality AI content and feel that their unique human experiences are being ignored. This situation shows that genuine human connection is now more valuable than ever because AI has made generic content too common. Businesses should focus on creating specific and heartfelt content made by actual people to stand out in this new landscape.
Sources
- AI Agents Are Becoming Enterprise Insiders. Security Models Need to Catch Up.
- Federal AI security guidance misses real-world enterprise threats
- Cisco taps AI, partners to modernize IT operations and fortify network security
- Britons are concerned about AI – but half still use it every week
- Akuna and Google open free AI filmmaking course in Africa
- New Research Reveals Gaps in AI Safeguards, Cyber Experts Discuss
- Francisco Lara-GarcĂa Discusses AI Border Surveillance on Univision | News
- Fort Bend County Judge race sparks controversy over AI endorsement video
- Several AI security market categories are headed for bigger platforms
- MD school board candidates view AI with trepidation
- Marissa Mayer Just Proved the AI Fatigue Backlash Is Real. Here’s What It Means for Your Brand
Comments
Please log in to post a comment.