The Newsroom: AI Infrastructure, Agent Security and Startup Funding Drive Technology Markets

The Newsroom · Technology Intelligence | September 24, 2026

The latest technology news across AI infrastructure, agent security, cybersecurity, semiconductors, enterprise software, startups, and funding.

The Newsroom · September 24, 2026

AI agent security moved from concept to infrastructure layer today as Okta and eleven other technology companies formally launched the Blueprint Alliance, a cross-industry coalition building a shared architecture for governing autonomous agents across enterprise systems.

The announcement anchored a day of significant developments across AI infrastructure, cybersecurity funding, semiconductor competition, and European defence technology. VAST Data introduced confidential AI capabilities for regulated industries. Delos Data emerged with $100 million to address AI networking bottlenecks. Palo Alto Networks launched a subscription service that puts frontier AI models to work hunting vulnerabilities. Cyera extended its Series G with $400 million. Axelera AI launched its second-generation European AI chip. Tekever raised $580 million for AI-powered autonomous defence systems. Markets turned lower as Treasury yields surged.

Lead Story

Lead Story

AI Agent Security Moves Into the Enterprise Infrastructure Layer

Source: Okta / Blueprint Alliance

Okta announced the formation of the Blueprint Alliance, a cross-industry coalition developing a shared architecture for securing AI agents across enterprise systems. Founding members include Amazon Web Services, CrowdStrike, Databricks, Docker, Google Cloud, Lovable, Okta, Proofpoint, Salesforce, ServiceNow, Wiz and Zscaler. GE Appliances and World Central Kitchen will serve as strategic advisers.

The alliance expands Okta’s existing secure agentic enterprise blueprint, first introduced in March 2026, into an open, multi-vendor reference architecture. The framework is organised around four questions: Where are my agents? What can they access? Who is accountable? How do I respond?

Okta simultaneously released new controls for AI agent authentication and delegated access. Three capabilities are generally available:

Capability	What It Does
Agent SSO	Extends Cross-App Access to all Okta single sign-on customers, replacing long-lived keys with short-lived tokens governed through identity policies.
Agent-to-Agent Connections	Sets rules for which agents can call one another and what each can access, recording the handoff in an auditable chain.
Resource Access Certifications	Provides recurring reviews of agent connections to help administrators remove excessive or unnecessary permissions.

Additional controls remain in development. Agent Gateway will sit between an agent and its tool calls to enforce policies and log interactions. Endpoint discovery will identify unmanaged agents running on employee devices. A Configuration Designer will visually map connections between agents and resources. Okta also plans to extend its kill switch to agents connected through the gateway, allowing active tokens to be revoked and running sessions terminated. Gateway and endpoint discovery are planned for Q3 availability, followed by Configuration Designer and the expanded kill switch in Q4.

The group’s reference architecture calls for each agent to have a distinct identity and an accountable human owner or operational team. Access should be limited to the task, delegation should remain traceable, and monitoring should continue while agents are running. Response options include token revocation, session termination, and controlled restoration of access.

What to note: Alliance members are building and testing cross-vendor interoperability using open protocols and security-signal standards. They plan to publish joint results and reference integrations so that a threat detected by one component can trigger a coordinated response across connected systems.

AI Watch

AI Watch 01

Enterprise AI Infrastructure Expands

VAST Data announced DataEnclave, a confidential AI capability within its VAST DataEngine, built on NVIDIA Confidential Computing. The system allows proprietary and open AI models to run against sensitive enterprise data inside customer data centres or trusted cloud hardware without exposing either customer data or model weights to infrastructure operators.

DataEnclave addresses a structural challenge in enterprise AI adoption: across financial services, healthcare, government and other highly regulated industries, valuable data remains inside tightly controlled environments where moving it to an external AI service is impractical or prohibited. Model builders face the inverse problem — they cannot distribute proprietary models into infrastructure they do not trust.

The system uses a hardware-isolated secure runtime and cryptographic attestation that verifies the environment and its enforced policy before sensitive assets are decrypted and loaded into the secure enclave container. Data remains protected in CPU and GPU memory during processing. Customer data keys remain under customer control, model keys and weights remain within the model builder’s trust domain, and infrastructure operators cannot access either.

Ecosystem partners include Cohere, CrowdStrike, Deepgram, Factory, Fundamental, NVIDIA and TwelveLabs. The system integrates with third-generation NVIDIA Confidential Computing across Hopper, Blackwell and Rubin architectures, with availability planned for Q1 2027 through VAST Data and participating OEM partners including Cisco and Supermicro.

VAST Data founder and CEO Renen Hallak framed the development as part of a broader shift: “Models are becoming a resource the operating system has to manage, the same way it manages data. That means knowing which model fits which task, what it can see, who can use it, and under what rules.”

AI Watch 02

AI Agents Are Changing Data-Center Architecture

Delos Data, a chip startup founded by veterans of Intel, raised over $100 million to develop chips and software that move data more efficiently inside AI data centres. The company is designing networking technology that works across different types of compute hardware rather than assuming every accelerator comes from a single vendor.

The early phase of the AI boom produced data centres made mostly of Nvidia GPUs wired together with proprietary networking technology. As the focus shifts from training AI systems to fielding agentic AI workloads, Nvidia rivals such as Cerebras Systems and AMD have gained traction, and Nvidia itself has begun offering multiple kinds of computing chips. Delos Data is building for that heterogeneous future.

“We don’t know what the next infrastructure architecture is going to be for agentic AI,” said Delos Data Chief Technology Officer and Co-Founder Dan Daly. “But we do know that we can provide the quickest, fastest way to move that data around.”

Pat Gelsinger, the former CEO of Intel who is now a partner with Playground, described the problem in operational terms: “I can have all the computers in the world, and if I can’t get these devices communicating effectively with one another, I’ve got lots of hot hardware, burning watts and sitting here twiddling their thumbs.”

Investors: Matrix Partners, Playground, Socratic Partners, Capricorn, Matter Venture Partners, IAG and DYNAMIQ.

Cybersecurity Watch

Cybersecurity Watch 01

Palo Alto Networks Expands AI-Powered Cyber Defense

Palo Alto Networks launched Continuous Frontier AI Defense, a subscription service that uses Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6-Cyber models, plus open-weight models, to hunt for security weaknesses in customer systems around the clock. The service is delivered through Unit 42, the company’s threat research and consulting arm.

After an initial scan of a customer’s entire estate, the service continuously tests web applications, APIs, cloud infrastructure, code repositories, and networks as they change. It attempts to prove that each weakness can be exploited, maps where an attacker could go next, and recommends code fixes or temporary virtual patches. Unit 42’s software decides which model handles each task, which the company says improves results and controls costs.

The service extends a one-off assessment Unit 42 introduced in April, which it began running on the two cyber models in August. According to Palo Alto Networks, that assessment found exposures at every customer it tested, with 37% rated high or critical severity. The company says it spent six months and $17 million developing the approach across more than 100 customer engagements. When it ran the service on its own systems, three weeks of testing surfaced as many exposures as it would normally expect to find in a year.

Palo Alto Networks says attackers using AI have cut the time from finding a flaw to exploiting it by almost 97% in some cases, from weeks to hours.

“AI has created an asymmetric advantage for threat actors against organisations trying to defend at human speed,” said Sam Rubin, Senior Vice President of Unit 42. “Modern cybersecurity requires machine-speed defense.”

Additional context: Anthropic reports that Claude Mythos has found more than 10,000 high-severity vulnerabilities in widely used software. Access to the model has been restricted; Anthropic briefly suspended access to its Mythos models in June to comply with US export controls, restoring it on 1 July after the controls were lifted. Pricing is based on an annual subscription and varies with the mix of models a customer chooses, and the service is available worldwide.

Cybersecurity Watch 02

Cyera Raises $400 Million

Source: Wall Street Journal

Cyera, the AI-native data and identity security company, raised $400 million from Growth Equity at Goldman Sachs Alternatives, extending the Series G that Evolution Equity Partners led in June. The valuation remains unchanged at over $12 billion.

The investment brings Cyera’s 2026 total to approximately $1.4 billion, following a $400 million round in January and a $600 million round in June. Total funding now exceeds $2 billion. The company was valued at $3 billion as recently as 2024.

The company said the new capital will fund three initiatives: an AI security roadmap, a push into the federal market, and growth across EMEA and APAC.

Cyera began by classifying enterprise data — determining what an organisation holds, how sensitive each piece is, and who can reach it. This year it extended that capability to the software acting on the data. Two products, Agent Guardian and Cyera Endpoint, track what AI agents do, logging tool calls, database queries, and actions in between, from cloud deployments down to endpoint devices. Customers named in the announcement include Paramount, Chipotle, and Valvoline.

Additional context: In July, Cyera acquired Oasis Security for approximately $1 billion. Oasis handles non-human identity — the credentials machines and agents use. Cyera CEO Yotam Segev wrote at the time that non-human identities inside Fortune 500 companies had grown nearly 500% in six months, calling them the fastest-growing identity type in the enterprise.

Semiconductor & AI Infrastructure Watch

Semiconductor Watch 01

European AI Chip Competition Expands

Axelera AI, the Netherlands-based AI acceleration company, launched its second-generation Europa architecture with a growing ecosystem of OEM and technology partners, including validated systems with Dell and Supermicro.

Europa is a purpose-built inference accelerator designed to extend Axelera’s edge-first architecture to more demanding enterprise and data-centre workloads. It is available in three configurations:

Form Factor	Target
Europa chip	Customers designing their own boards.
Axelera Edge 232p	Half-height, half-length PCIe card.
Axelera Server 250p	Full-height, full-length PCIe card.
Titania (future)	Chiplet architecture targeting data centres and supercomputers.

The company reports a sales pipeline exceeding $1.5 billion and deployments with more than 600 customers across defence, robotics, drones, retail and security applications. Axelera claims up to 6x more tokens per second per watt than GPU-based solutions for certain workloads.

The launch reflects a broader shift in AI infrastructure economics. Inference is recognised as the bulk of AI compute load, but rising cloud costs, exponential token usage, power and GPU constraints, and increasingly capable smaller models are changing where AI can and should run. Many enterprise applications do not require the largest frontier models or the most expensive hardware. They need infrastructure designed to run persistent AI workloads securely and at a cost organisations can sustain.

Market context: Axelera’s customers span AI factories and gigafactories in Europe, with particular relevance in markets where security, compliance and data sovereignty requirements limit the use of public cloud AI — including financial services, healthcare, legal, defence and government.

AI Infrastructure Watch 02

AI Networking Becomes a Bigger Infrastructure Layer

The Delos Data funding and Axelera’s Europa launch together illustrate a factual development in AI infrastructure: investment is expanding beyond GPUs into networking, data movement, and system-level architecture.

As AI workloads diversify across different accelerators and deployment environments, the ability to move data efficiently between heterogeneous compute resources becomes a distinct infrastructure requirement. Delos Data is designing for that heterogeneity. Axelera is designing for inference at the edge and in enterprise environments. Both address the same underlying challenge: AI infrastructure is no longer a single-component problem.

Factual development: AI infrastructure is expanding beyond GPUs into networking, data movement, and system-level architecture.

Startup & Funding Watch

Startup & Funding Watch 01

Tekever Raises $580 Million

Source: Reuters

European autonomous-systems developer Tekever raised $580 million in the first close of a Series D financing round, valuing the Portuguese-British company at $6.4 billion. UC Investments and Baillie Gifford led the round, with participation from Merlyn Advisors, Crescent Cove, Ventura Capital and Iberis Capital.

Tekever creates AI-driven autonomous systems, including surveillance drones. The company stated that its drones have accumulated over 50,000 operational flight hours in Ukraine since 2022. In July, the UK Ministry of Defence selected Tekever to supply surveillance technology under a 10-year programme valued at up to £400 million (approximately $530 million).

“Artificial intelligence, autonomy and sovereign technological capability will be fundamental to the security and resilience of democratic societies in the decades ahead,” said Ricardo Mendes, Tekever’s founder and CEO. “This investment gives us the resources to move faster, scale further and continue building a global technology leader from Europe.”

Tekever plans to use the funds to strengthen its global footprint, enhance industrial and technological capacities, and accelerate strategic acquisitions.

European defence tech context: The round adds Tekever to a growing list of European defence technology startups securing large funding rounds. German autonomous drone companies Helsing and Quantum Systems raised $1.8 billion and €1 billion respectively in July. British air and missile defence firm Cambridge Aerospace secured $300 million in August, and German AI-powered unmanned systems startup Stark raised €500 million in June.

Enterprise Software Watch

Enterprise Software Watch 01

AI Agent Governance Becomes Enterprise Software Infrastructure

The Blueprint Alliance establishes a direct bridge between enterprise software, identity management, cybersecurity, and AI agents. The participation of AWS, Salesforce, ServiceNow, Databricks, Okta, CrowdStrike, Wiz, and Zscaler signals that agent governance is being treated as shared infrastructure rather than a single-vendor problem.

The alliance’s reference architecture treats every agent as a first-class identity with a distinct owner, limited access, traceable delegation, continuous monitoring, and defined response options. Members are building and testing cross-vendor interoperability using open protocols and security-signal standards, with plans to publish joint results and reference integrations.

Okta previously extended its agent governance platform to Amazon Bedrock and third-party identity providers. Separate industry work has focused on identifying the people behind agent actions, including Proof’s participation in FIDO Alliance agent authentication standards.

Why it’s in the Newsroom: The formation of the alliance reflects a practical reality — AI agents access multiple systems, and securing them requires coordination across the identity, cloud, cybersecurity, data and software layers that agents traverse.

Technology Markets

AI / Semiconductor Market Reaction — September 24, 2026

U.S. markets fell across the board on September 23 (local time), ending the Nasdaq’s three-day record-setting run. The downturn was triggered after S&P Global’s September manufacturing and services indexes surged to their highest levels in about five years while inflationary pressures intensified. Federal Reserve Governor Michael S. Barr stressed the need for another rate hike in October, and U.S. Treasury yields rose sharply across the board.

Major Indexes

Index	Close	Change
Dow Jones Industrial Average	51,511.77	−351.92 (−0.68%)
S&P 500	7,706.03	−58.61 (−0.75%)
Nasdaq Composite	26,936.04	−308.24 (−1.13%)
Nasdaq-100	30,470.29	−262.10 (−0.85%)
Cboe Volatility Index (VIX)	15.24	+1.03 (+7.24%)

Semiconductor Stocks

Semiconductor stocks, which had been climbing, turned lower as Treasury yields fluctuated and expectations of additional Fed rate hikes strengthened.

Ticker	Close	Change
NVIDIA	$225.51	−$3.36 (−1.47%)
Micron Technology	$1,071.88	−$24.28 (−2.22%)
SK hynix ADRs	$189.28	−$6.09 (−3.12%)
SanDisk	$1,816.57	−$70.47 (−3.73%)
Philadelphia Semiconductor Index (SOX)	12,534.27	−155.55 (−1.23%)
Direxion Daily Semiconductor Bull 3X (SOXL)	$146.25	−$5.70 (−3.75%)

Treasury yields: The 10-year Treasury yield surged 0.135 percentage points to 5.102%, topping 5.4% intraday — its highest level in 19 years since mid-2007. The two-year yield climbed to 4.895%, and the 30-year yield rose to 5.392%.

Significant Movers

Palantir Technologies	up approximately 3%
Alphabet	$339.34 (−3.37%)

Data note: All index, price, and yield figures are reported market data. Percentage changes are CODEW calculations from reported closing prices. Market data is as of the September 23, 2026 U.S. close.

What to Watch

1. AI-agent security standards	The Blueprint Alliance plans to publish joint results and reference integrations. Watch for the first cross-vendor interoperability demonstrations and whether additional identity, cloud, and security vendors join the coalition.
2. Enterprise confidential AI deployments	VAST DataEnclave is scheduled for Q1 2027 availability through VAST Data and OEM partners including Cisco and Supermicro. Watch for early customer deployments in financial services, healthcare, and government.
3. AI networking and data-movement infrastructure	Delos Data’s $100 million raise signals investor interest in solving AI interconnect bottlenecks. Watch for product timelines, customer announcements, and whether other networking startups attract comparable funding.
4. Cybersecurity funding	Cyera’s $400 million extension brings its 2026 total to $1.4 billion. Watch whether other data-security and AI-agent-security companies attract similar late-stage capital, and whether the federal market push produces contract announcements.
5. European AI-chip and autonomous-systems investment	Axelera AI’s Europa launch and Tekever’s $580 million raise reflect distinct but related trends — European semiconductor sovereignty and European defence technology investment. Watch for Axelera’s Titania chiplet architecture timeline and Tekever’s acquisition activity following its Series D.

Related CODEW Coverage

→ Daily Tech Briefing — Why the Blueprint Alliance Matters for Enterprise Identity
→ AI Watch — Confidential AI and the Future of Regulated Data
→ AI Infrastructure Watch — Networking Becomes the Next AI Bottleneck
→ Cybersecurity Watch — AI-Powered Defence and the Speed Problem
→ Semiconductor Watch — Europe’s AI Inference Ambitions
→ Startup Funding Watch — Defence Tech’s Funding Surge
→ Enterprise Software Watch — Agent Governance as Shared Infrastructure
→ Networking Watch — Data Movement in Heterogeneous AI Data Centres

Next in the Newsroom

The September 25 edition will cover developments in AI model releases, cloud infrastructure earnings previews, and enterprise software M&A activity.

The CODEW Stat

12 alliance members · $1.98B in disclosed funding · 5 developments to watch Okta’s Blueprint Alliance launched with twelve founding members spanning identity, cloud, cybersecurity, data and enterprise software — treating AI agent governance as shared infrastructure rather than a single-vendor problem. Across the day’s coverage, disclosed funding totalled roughly $1.98 billion across Delos Data ($100M), Cyera ($400M), Tekever ($580M) and previously reported rounds, while VAST Data, Palo Alto Networks and Axelera AI advanced enterprise AI infrastructure, AI-powered cyber defence and European AI inference, respectively. Markets closed lower as the 10-year Treasury yield topped 5.4% intraday for the first time in 19 years.

THE CODEW · TECHNOLOGY INTELLIGENCE

Editorial Note

The Newsroom reports what happened. Strategic interpretation and competitive analysis are reserved for the Daily Tech Briefing and Watch Tech Series. This edition covers AI infrastructure, agent security, cybersecurity, semiconductors, enterprise software, startups, and funding for September 24, 2026.

Market data is reported as of the September 23, 2026 U.S. close. Percentage changes are CODEW calculations from reported closing prices. Educational content only. Not investment advice. Analysis is based on company announcements, official product disclosures, investor relations releases, and reporting from Reuters, the Financial Times, the Wall Street Journal, Bloomberg, and CNBC where appropriate. Some products referenced may be affiliate partners — see our Affiliate Disclosure for full details. Platform coverage, data sources, and methodologies can change as the intelligence platform evolves.