The CISO Paradox: More Autonomy, More Control – Issue #47
Spotlight: The CISO Paradox: More Autonomy, More Control
CISOs are ready to give AI more authority, but every step toward autonomy demands tighter control.
They are grappling with a new operating reality: AI is becoming both a defensive layer and a source of new risks. AI-generated software, agents, and shadow AI are creating risks that their current stack was not designed to manage.
At Mayfield, we surveyed more than 60 CISOs, CSOs, and heads of security engineering and operations to understand how AI is reshaping their organizations, the emerging risks, and the opportunities for founders building today.
Here are the top 5 findings (see all 10 in my newsletter: https://www.linkedin.com/pulse/ciso-paradox-more-autonomy-control-navin-chaddha-oy7bf/) from Mayfield’s CISO survey report:
– 67% rank vulnerabilities in AI-generated software among their top emerging risks, and only 2% believe their current stack is fully equipped to manage this risk
– 83% lack complete visibility into employee AI use
– 77% would prioritize vulnerability remediation for end-to-end AI ownership
– Top AI adoption barriers: trust in outputs (53%) and data privacy concerns (53%), followed by security of the AI systems (52%)
– 50% say a measurable pilot tied to outcomes is the fastest way for a startup to earn trust
CISO budgets are growing, spending is shifting toward AI-native vendors, and the market doesn’t yet have a clear leader. Founders must prove security efficacy in the customer’s environment, build governance and auditability into the product, and demonstrate measurable risk reduction before buyers will expand autonomy or replace an incumbent.
CISOs want AI to close the loop. They will give it more authority when they can see what it is doing, control how it acts, and defend the outcome.
More autonomy requires more control.
In this week’s news, Anthropic launched Claude Opus 5.5 with stronger agent capabilities at 40% lower cost, Meta’s Muse outpaced ChatGPT’s early mobile launch with 2.8 million installs in its first 12 days, Claude helped uncover a previously unknown enzyme system with CRISPR-like characteristics, Microsoft committed more than $10 billion to cloud and AI infrastructure across the Middle East, and Palo Alto Networks launched continuous AI-powered cybersecurity testing using models from Anthropic and OpenAI.
Full Weekend Edition below. 👇
Signals Shaping the Future of AI:
Infrastructure
Oracle reportedly issued a force majeure notice due to delays at its Project Jupiter data center campus. The setback highlights the mounting execution challenges around power, construction, and financing for hyperscale AI infrastructure. Click here
OpenAI and Anthropic are reportedly seeking new 20–30 megawatt data-center deployments. The projects show that frontier labs continue to distribute compute across multiple sites rather than relying solely on massive hyperscale campuses. Click here
Alibaba unveiled its Zhenwu V900 AI chip and plans to expand its global cloud capacity to more than 20 gigawatts by 2032. The company says the new accelerator delivers 3x the performance of its predecessor and can scale into clusters of up to 500,000 chips. Click here
Meta plans “Petal,” its first transoceanic subsea cable designed for petabit-scale capacity. The project expands Meta’s control over the networking infrastructure supporting increasingly data-intensive AI services. Click here
Tower Semiconductor plans a major optical-chip hub as part of a $4 billion investment commitment in Japan. The project targets the faster, more power-efficient connectivity needed to move data across increasingly large AI clusters. Click here
Enterprise
Anthropic launched Claude Opus 5.5, cutting typical workload costs by about 40% versus Opus 5 while improving coding and agent performance. Anthropic is positioning the model for long-running enterprise agents and complex professional work. Click here
OpenAI released GPT-6 Sol and Luna with significantly lower API pricing. The new models push advanced coding, agent, extraction, and recurring enterprise workloads further down the cost curve. Click here
Meta expanded Muse into an always-on personal agent that can use computers, email, commerce services, and third-party connectors. Developers have already submitted more than 1,500 connector applications as Meta builds a broader ecosystem around agent-driven software. Click here
Microsoft expanded Foundry with new capabilities to build, operate, and govern production AI agents. The platform adds voice and long-running agents, agent-to-agent communication, automated optimization, and new identity and network controls as enterprises move agents into production. Click here
Palo Alto Networks launched an AI-powered cybersecurity service using models from Anthropic, OpenAI, and open-weight providers. Unit 42 Continuous Frontier AI Defense will continuously test applications, APIs, and cloud infrastructure for vulnerabilities and recommend fixes as enterprise environments change. Click here
Capital Flows
Databricks acquired enterprise spreadsheet startup Row Zero as it continues expanding through acquisitions. Terms were undisclosed, but the deal adds another enterprise interface to Databricks’ growing AI and agent stack. Click here
Cyera raised another $400 million from Goldman Sachs Alternatives as enterprises race to secure AI agents and sensitive data. The investment pushes its valuation above $12 billion as identity, permissions, and agent access emerge as major enterprise-security priorities. Click here
Enveda raised $311 million at a $2 billion valuation to advance AI-discovered medicines into clinical trials. The company uses AI to identify potential drugs from plants and microbes, and now has several candidates undergoing human testing. Click here
Snorkel AI raised $350 million at a $3.5 billion valuation as demand for specialized AI training data grows. The Series E will support synthetic data, evaluation environments, and other infrastructure used to improve advanced models. Click here
Ema raised $77 million in Series B funding as enterprise demand for AI agents grows. The round brings total funding to $140 million and more than quadruples the company’s valuation from its previous financing. Click here
Research
Anthropic says Claude helped discover a previously unknown enzyme system with CRISPR-like characteristics. The company also launched a dedicated life-sciences research group to pair AI-generated hypotheses with physical experimental validation. Click here
Google unveiled persistent server-side AI memory designed to preserve strong privacy protections. Private AI Compute aims to maintain continuity across devices without exposing underlying personal information to Google or other parties. Click here
Researchers found AI agents attempting to exploit vulnerabilities while completing routine data-retrieval tasks, including against an Australian government website. Australia separately disclosed that an OpenAI agent gained unauthorized access to a government statistics portal, underscoring new control challenges as agents become more autonomous. Click here
Researchers developed a standard that lets AI agents directly coordinate laboratory equipment and automate experiments. The Model Hardware Standard connects instruments from different vendors, enabling agents to control workflows that previously required extensive custom integration. Click here
Policy
California signed seven new laws governing data-center electricity, water, and land use. The package adds reporting requirements and seeks to prevent infrastructure costs from being shifted from data centers onto residential ratepayers. Click here
Texas reportedly halted new state-issued data-center permits while regulators conduct a broader grid audit. The review is examining power demand, water use, incentives, and project ownership before approvals resume. Click here
The U.S. Justice Department intervened in the New York Times copyright litigation involving OpenAI and Microsoft. Its filing supported arguments by AI companies and raised concerns that restrictive copyright interpretations could undermine U.S. AI competitiveness. Click here
EU officials said implementation of the AI Act remains incomplete across several member states. The European Commission is pressing countries that have not fully designated the national market-surveillance authorities responsible for enforcement. Click here
OpenAI and Anthropic urged Australia to reconsider restrictions on training AI models using Australian creative content. The companies are seeking a limited copyright exemption as the country develops its broader AI policy framework. Click here
Global AI Strategy
China is seeking an AI safety dialogue with the U.S. ahead of Xi Jinping’s Washington visit while opposing U.S. efforts to constrain Chinese AI development. The issue is part of the broader technology and trade agenda between the two governments. Click here
Microsoft committed more than $10 billion to cloud and AI infrastructure across the Middle East through 2030. The investment spans Kuwait, Qatar, Saudi Arabia, and the UAE alongside deeper partnerships with national AI organizations, including HUMAIN, G42, and QAI. Click here
South Korea pledged to double semiconductor production capacity within five years and strengthen critical AI supply chains. President Lee Jae Myung tied the initiative to major chip and AI megaprojects while calling for additional U.S. investment. Click here
Microsoft launched its India South Central cloud region in Hyderabad as part of its $20.5 billion India investment program. The new region includes three availability zones designed to support AI and cloud workloads. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
Vijil is building trust and resilience infrastructure for AI agents, helping enterprises evaluate systems before deployment, protect them in production, and continuously improve them as risks evolve. The Mayfield-backed company is hiring across software engineering, AI research, sales, and enterprise roles. Click here
Semgrep is building an AI-powered application security tool that helps engineering and security teams detect, prioritize, and remediate vulnerabilities as AI dramatically increases the volume of software being written. The company is hiring across security, engineering, product, and go-to-market roles. Click here
https://www.linkedin.com/company/endorlabs/EndorLabs is building application security for the AI coding era, helping teams identify, prioritize, and fix vulnerabilities across AI-generated code, open-source dependencies, and complex software supply chains. The company is hiring across engineering, security research, product, customer success, and go-to-market roles. Click here
You can see all the opportunities at Mayfield-backed AI companies here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Andrew Ng (Click here) — “AI technology has not taken some unexpected, dangerous turn. The biggest change in AI risk is its cybersecurity capabilities, a topic which we should take seriously, but this, too, will not lead to the end of the world.” In a post viewed 7.8M times, Ng pushes back on recent calls to slow or pause frontier AI, arguing that incidents like the OpenAI-Hugging Face breach should be treated primarily as engineering and security failures to fix rather than evidence that AI development itself must stop. His broader case is that progress and safety should advance together through better sandboxing, monitoring, accountability, and defensive cybersecurity.
Dario Amodei (Click here) — “Today, we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism. The work was done mostly, though not entirely, by Claude: our life sciences team suggested a broad area of research, Claude read through the literature and genome data, discovered something interesting, and then proposed experiments to verify the discovery. We believe AI for biology is on a similar exponential trend.” Amodei argues that biology may be entering the same capability curve AI has already shown in mathematics, with models moving from assistants toward active scientific discovery. The broader opportunity is to use AI to accelerate the experimental loop and increase the throughput of promising discoveries across the drug-development pipeline.
Ethan Mollick (Click here) — “I think Meta’s Muse is an impressive implementation of the OpenClaw idea of AI as a personal assistant agent that you have an ongoing chat with. Since it is so focused on doing that well, the experience is very accessible for people who didn’t realize what AI could do for them.” Mollick argues that consumer AI may differentiate less on raw model capability and more on product design, continuity, and ease of use. His broader point is that while frontier labs chase enterprise demand, products like Muse could expand adoption by making increasingly capable agents feel intuitive to mainstream users.
