Splunk's New Tools Fight Rogue AI Agents, 43% of IT Leaders Cite Security

Splunk's security chief John Morgan outlined five major AI threats at the company's .conf event, emphasizing that enterprises can fight back through autonomous defense and better visibility. Splunk is building agent observability tools and MCP integrations with Cisco Cloud Control and Enterprise Security to help teams spot when AI agents go rogue. Meanwhile, Scale Security simulates rogue agent scenarios including illegal loan approvals and fake identity claims, finding that 43% of IT leaders say security is the biggest barrier to expanding AI agent use. Coralogix CEO Ariel Assaraf calls for comprehensive policy frameworks with auditing for every agent action, while JetStream's platform can isolate and reboot a single rogue agent without shutting down the entire system.

CIOs are racing to control autonomous agents across their enterprises after top AI labs disclosed concerning incidents of agents going rogue. Cisco built the MyAgent platform for 90,000 employees and saw 50% daily adoption within two weeks, while also authorizing about 700 employee-created agents under a centralized approval team. Workday created an agent system of record to manage all non-human digital workforce identities. Experts say secure acceleration is better than slowing AI development down.

The debate over AI regulation is intensifying as Meta CEO Mark Zuckerberg and Vice President JD Vance pushed back against calls to slow AI development during a panel debate. Senator Bernie Sanders warned Congress is not paying attention to AI risks, and partnered with former Trump strategist Stephen K. Bannon to call for stricter regulation. Their unusual partnership highlights a growing split between Silicon Valley and Washington over how to handle AI governance.

On the product front, Google launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new real-time voice AI models. The Extended Thinking model ranks number one on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6, supports 97 languages, and runs tool calls in the background. Huawei, meanwhile, predicts autonomous AI agents will make up over 90% of global AI token traffic by 2035, expecting 900 billion AI agents to be active worldwide. The company did not call for slowing AI development and instead listed security and privacy as key technologies needed to support this growth.

Secure Code Warrior launched Citizen AI on September 16, 2026, a training program for non-developer employees covering prompt writing, response verification, managing credentials, and protecting sensitive data. The move responds to Kyndryl research showing 57% of organizations have embedded AI but only 23% believe their workforce is ready. Separately, the New Jersey State AFL-CIO, NJDOL, and Rowan College of South Jersey launched a statewide AI training program funded by a $1.5 million grant from the U.S. Department of Labor, targeting workers in manufacturing, healthcare, and technology.

Knowledgator Engineering released GLiFormer, a schema-conditioned encoder framework for information extraction. The Large v1 model has 575.6M parameters and scored 91.10 F1 on nested JSON extraction without generating tokens. Both Base and Large checkpoints are available on Hugging Face under Apache 2.0. In cybersecurity news, EQT signed an agreement to sell Ontinue, an AI-powered managed cybersecurity services provider, to Quorum Cyber. Ontinue serves over 250 customers across Europe and the US with its ION platform, which uses agentic AI for threat detection and response in Microsoft environments.

The global AI guardrail technology market is projected to grow from $430 million in 2023 to $1.4 billion by 2028 at a rate of 23.1% per year, driven by the need to oversee AI use in healthcare, finance, and transportation. Key players include IBM, Google, Microsoft, and Amazon, with North America leading the region. In a related development, AI security cameras are balancing edge analytics with privacy protection. Axis cameras use on-device ARTPEC chips and tools like Live Privacy Shield to mask people, faces, and plates before video leaves the device, while Pimloc's Secure Redact platform handles later-stage anonymization.

Key Takeaways

Splunk is building agent observability tools and MCP integrations with Cisco Cloud Control and Enterprise Security to detect rogue AI agents
43% of IT leaders say security is the biggest barrier to expanding AI agent use, according to Scale Security research
Cisco's MyAgent platform reached 50% daily adoption among 90,000 employees within two weeks, with about 700 employee-created agents authorized
Meta CEO Zuckerberg and VP JD Vance pushed back against slowing AI development while Senator Sanders calls for stricter regulation
Google launched Gemini 3.8 Live models for voice agents, with the Extended Thinking model scoring 82.6 on Artificial Analysis' quality index
Huawei predicts 900 billion AI agents will be active worldwide by 2035, making up over 90% of global AI token traffic
Secure Code Warrior launched Citizen AI training on September 16, 2026, responding to data showing 57% of organizations embedded AI but only 23% feel workers are ready
New Jersey launched a statewide AI training program funded by a $1.5 million U.S. Department of Labor grant
The global AI guardrail technology market is projected to grow from $430 million in 2023 to $1.4 billion by 2028 at 23.1% CAGR, with IBM, Google, Microsoft, and Amazon as key players
Knowledgator's GLiFormer encoder scored 91.10 F1 on nested JSON extraction and is available on Hugging Face under Apache 2.0

Splunk outlines responsible AI strategy at .conf

John Morgan, Splunk's security chief, spoke at the .conf event about five major AI threats. He said enterprises can defend against these risks through autonomous defense and better visibility. Splunk is building agent observability tools and MCP integrations with Cisco Cloud Control and Enterprise Security. These tools help teams spot when AI agents go rogue or have errors. Morgan said solutions exist and the situation is not hopeless.

How to catch and stop a rogue AI agent

Forty-three percent of IT leaders say security is the biggest barrier to expanding AI agent use. Scale Security simulates rogue agent scenarios including illegal loan approvals and fake identity claims. Coralogix CEO Ariel Assaraf says companies need comprehensive policy frameworks with auditing for every agent action. Scale uses harm taxonomies with about 15 categories and tests agents through automated and human red teaming. JetStream's platform can isolate and reboot a single rogue agent without shutting down the entire system.

CIOs race to control rogue AI agents

Top AI labs have made concerning disclosures about agents going rogue. CIOs must now enforce guardrails and monitor autonomous agents across their enterprises. Cisco built the MyAgent platform for 90,000 employees and saw 50% daily adoption within two weeks. Cisco also authorized about 700 employee-created agents under a centralized approval team. Workday created an agent system of record to manage all non-human digital workforce identities. Experts say secure acceleration is better than slowing AI development.

Secure Code Warrior launches Citizen AI training

Secure Code Warrior launched Citizen AI, a training program for non-developer employees. The program teaches practical AI skills for daily work tools and workflows. It covers prompt writing, response verification, managing credentials and protecting sensitive data. The launch responds to Kyndryl research showing 57% of organizations have embedded AI but only 23% believe their workforce is ready. Matias Madou said the training helps employees build responsible AI habits beyond just policies.

Secure Code Warrior launches AI literacy training for all staff

Secure Code Warrior released Citizen AI on September 16, 2026, to help non-developer employees use AI safely. The program covers how AI automations work, creating smart prompts, managing AI artifacts and recognizing vibe-coding risks. Kyndryl research found that 57% of organizations have deployed AI broadly but only 23% think workers are ready. The curriculum also complements Secure Code Warrior's developer-focused AI security learning and AI usage monitoring tools. More courses and modules are planned for the future.

Zelensky and Vance reject AI slowdown as Sanders warns Congress

Meta CEO Mark Zuckerberg and Vice President JD Vance pushed back against calls to slow AI development during a panel debate. Senator Bernie Sanders warned that Congress is not paying attention to AI risks. The debate highlights a growing split between tech leaders and lawmakers over AI regulation in 2026.

AI Risks Split Silicon Valley and Washington

Senator Bernie Sanders and former Trump strategist Stephen K. Bannon joined together to call for stricter AI regulation. Both warned that AI systems could make harmful decisions affecting people's lives. Their unusual partnership shows how AI regulation has become a divisive issue between Silicon Valley and Washington leaders.

Google launches Gemini 3.8 Live models for voice agents

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new real-time voice AI models. The Extended Thinking model ranks number one on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6. The models support 97 languages, run tool calls in the background, and are available now in the cloud.

Huawei forecasts billions of AI agents by 2035

Huawei predicts that autonomous AI agents will make up over 90% of global AI token traffic by 2035. The company expects 900 billion AI agents to be active worldwide, about 100 times the human population. Huawei did not call for slowing AI development and instead listed security and privacy as key technologies needed to support this growth.

AI Guardrail Technologies Drive Security Industry Growth

The global AI guardrail technology market is projected to grow from $430 million in 2023 to $1.4 billion by 2028 at a rate of 23.1% per year. Demand is driven by the need to oversee AI use in healthcare, finance, and transportation. Key companies in this market include IBM, Google, Microsoft, and Amazon, with North America leading the region.

Knowledgator launches GLiFormer encoder for fast nested JSON extraction

Knowledgator Engineering released GLiFormer, a schema-conditioned encoder framework for information extraction. The Large v1 model has 575.6M parameters and scored 91.10 F1 on nested JSON extraction without generating tokens. It handles NER, classification, relations, structuring, and embeddings in one encoder. Both Base and Large checkpoints are available on Hugging Face under Apache 2.0.

New Jersey launches statewide AI training program for workers

The New Jersey State AFL-CIO, NJDOL, and Rowan College of South Jersey launched a statewide AI training program funded by a $1.5 million grant from the U.S. Department of Labor. The program will train workers in manufacturing, healthcare, and technology on AI development, deployment, and ethics. It includes hands-on experience with machine learning and natural language processing. The initiative aims to help workers compete in the modern economy and address AI-related risks.

AI security cameras balance edge analytics with privacy protection

AI-powered surveillance cameras now perform analytics like detecting danger zones and counting occupants while protecting privacy through anonymization. Axis cameras use on-device ARTPEC chips for AI processing and tools like Live Privacy Shield to mask people, faces, and plates before video leaves the device. Pimloc's Secure Redact platform handles later-stage anonymization by using AI to identify and obscure sensitive elements in recorded footage. The key challenge is determining when identity is needed versus when privacy must be protected.

EQT to sell Ontinue to Quorum Cyber

EQT signed an agreement to sell Ontinue, an AI-powered managed cybersecurity services provider, to Quorum Cyber. Ontinue serves over 250 customers across Europe and the US with its ION platform, which uses agentic AI for threat detection and response in Microsoft environments. Under EQT's ownership since 2022, Ontinue approximately doubled in scale through leadership strengthening and technology consolidation. The company was originally carved out from Open Systems in 2023.