Daily · AI Safety, Security, and Sovereignty · October 11, 2026

Key points

AI Agents Execute Full OT Attack Chains

Booz Allen Hamilton’s operational technology lab tested eight scenarios examining advanced AI models' capabilities in autonomous attack chains. The models achieved objectives across all eight scenarios, demonstrating that frontier AI agents can execute full OT attacks at machine speed.

In one test, an agent probed the network for common robotic protocols, identified a robot, discovered its API, gained administrative access, mapped protection zones and motion limits, and moved the arm in minutes. In another, the models progressed from a perimeter compromise to actions inside an industrial control network in just over 16 minutes. The test environment included PLCs, HMIs, a SCADA platform, a VFD, a robotic arm, and sensors, modeled after a general manufacturing facility.

The models were not given source code or engineering documents; agents had to wait for human approval before exploiting security issues or taking actions with physical impact. In the SCADA test, the model determined that the gateway exposed live, pre-auth connections to 14 OT devices, meaning compromising one device provided access to 14 others. Booz Allen declined to identify the specific models, describing them as two of the latest frontier models from leading providers.

The report notes that OpenAI, Anthropic, and Google have announced initiatives to give critical infrastructure owners access to their advanced models to defend against agentic attacks. This suggests a shift in risk management where the bottleneck moves from detection to containment and human oversight of autonomous actions.

Korean Banks Boost Cyber Budgets

Shinhan Bank reported on October 1 that personal information of about 25,000 customers, including names, phone numbers, and annual incomes, was leaked in a hacking attack reportedly using advanced AI tools. Similar leaks or exposure of customer information were also reported at Hana Bank and KB Kookmin Bank this month.

In response, KB Kookmin Bank plans to increase its cybersecurity budget to more than 100 billion won (US$74 million) in 2027, up from 86.07 billion won in 2026. Shinhan Bank plans to allocate a record cybersecurity budget, and Hana Bank plans to invest tens of billions of won in strengthening security systems next year.

The Financial Supervisory Service (FSS) identified Internet Protocol addresses in 12 countries suspected of being linked to the AI-assisted hacking attacks, though investigators were unable to determine the locations linked to some of the IP addresses. These moves follow President Lee Jae Myung's call over the weekend for a thorough investigation into the attacks.

Open Source Verification Bottleneck

Anthropic’s OSS Scanner surfaced more than 29,000 candidate vulnerabilities in open-source projects over six months. Its human review pipeline, which relies on six external security research firms to reproduce and triage findings, has worked through only about 6,000. As of October 2, outside security firms had confirmed 5,674 of the 6,123 findings they reviewed as valid, but only 516 vulnerabilities had been patched upstream.

Anthropic sent 6,157 findings to maintainers, and 584 CVE and GitHub Security Advisory identifiers were issued. Roughly 23,000 candidates remain unreviewed. Expert penetration testers checked 97 critical and high-severity findings from an early version of the scanner across 48 projects: 85 cleared the bar for disclosure, 11 duplicated known issues, and only one was a false positive.

Todd Ouska, founder of wolfSSL, reported that 72 of the 74 reports his team received were valid and five became CVEs. Daniel Stenberg, founder of The cURL Project, said OSS Scanner found multiple issues in curl, including one of the worst vulnerabilities the project has seen in recent years; curl had shut down its bug bounty in January after a flood of AI-generated reports. Anthropic is giving eligible projects free, periodic scans powered by its top models, with findings going directly to maintainers without prior validation.

Capital Flows and Model Sovereignty

DeepSeek is nearing a second external funding round of at least 80 billion yuan (~$12B) and is considering expanding it to as much as 100 billion yuan (~$15B). The round is priced at a ~500 billion yuan (~$75B) valuation ahead of a possible early-2027 Shanghai STAR Market IPO. Tencent and CATL are among the largest backers alongside Geely Auto, Monolith Management, and Loyal Valley Capital.

Mistral launched a preview of Mistral Large 4, a 1-trillion-parameter natively multimodal open-weight model with 49B active parameters. It was trained on 3,800 NVIDIA Grace Blackwell GPUs. The preview API is live now, with weights scheduled for release by the end of the month. Mistral reports 61.7% on DeepSWE v1.1 for coding and agentic tasks and the top open-weight score outside China on the Artificial Analysis Cyber Index.

TypeSafe AI raised about $870 million in a Series A at a $7.5 billion valuation led by Andreessen Horowitz, with participation from Sequoia Capital and DCVC. This comes less than four weeks after TypeSafe left stealth with a $40 million seed. TypeSafe says its Jev model is used by about a third of the Fortune 500.

OpenAI published 722 AI-generated math manuscripts produced by an unreleased internal frontier model, with about 300 having Lean formalizations. The average result used about three hours of ChatGPT Pro-equivalent compute. A day later, OpenAI withdrew three manuscripts over a sign error and revised 14 others. Mathematicians note that with the model unreleased, outsiders cannot replicate the results.

Regulatory Frameworks and Safety Controls

California Governor Gavin Newsom signed two bills establishing a first-in-the-nation framework for Independent Verification Organizations (IVOs) and a state registry for AI auditors. Anthropic supported both bills in August; OpenAI formally endorsed them the same day Newsom signed them into law. OpenAI global affairs chief Chris Lehane wrote that while they prefer federal requirements, California can help establish the rules of the road in the absence of federal action.

Microsoft CEO Satya Nadella called for advanced AI systems to be built with containment, independent controls, and a human-controlled emergency brake. He stated that treating frontier models like insider risks is a way to build such a system. His comments come amid warnings from Bill Gates, Dario Amodei, Sam Altman, and Elon Musk about insufficient AI safety protocols.

President Trump presented a voluntary one-page AI safety accord at a White House luncheon, signed by top executives of Anthropic, Google, Meta, OpenAI, SpaceX, and Nvidia. The document encourages signees to work with independent external auditors. Brown University professor Suresh Venkatasubramanian called it a performance of an attempt to show action when in fact no action actually happened.

METR raised approximately $71 million in commitments over six months, up from total 2024 contributions of $13.6 million. Vals AI announced a $40 million funding round. Anthropic announced it will embed employees from Faculty and Accenture to test model safeguards, with contracts giving them the right to publish key findings.

Geopolitical Tech and Infrastructure Risks

Elon Musk vowed to price Starlink affordably in India, escalating his public feud with Mukesh Ambani days before Jio's record IPO. Musk posted on X that if Starlink is too expensive, no one will buy it, so SpaceX will need to make pricing affordable. He accused Modi's government and Ambani of stalling Starlink's entry to favor Jio; India's government denied any bias, noting all three satellite ventures seeking to launch are awaiting final security clearance. Jio Platforms is targeting a valuation of $106bn in India's largest-ever IPO.

Yandex reported gradual restoration of access to its services after damage to the company's data center infrastructure in Vladimir, with some functions still unavailable. The Vladimir data center is the third Yandex infrastructure site damaged by drone attacks in recent days, following data centers in Ryazan and Kaluga regions. Regional governor Alexander Avdeyev reported power supply in Vladimir restored to 80% while heat and water supply were fully restored.

Shing-Tung Yau launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence in Beijing, uniting eight institutions to build AI models for mathematical research. The alliance aims to create a network of more than 1,000 scientists. Yau said the gap between China and the United States in AI systems capable of advancing mathematics and physics is very large.

Iran and Russia discussed expanded ICT cooperation in AI and cybersecurity at the 3rd Russia-Iran ICT Forum held in Moscow. As many as 760 delegates from 120 countries took part in the Global Digital Forum. In December 2025, top Iranian and Russian tech companies signed five MOUs on AI, cybersecurity, smart government, blockchain, and fintech.

Our read

The convergence of autonomous AI capabilities in security research and operational technology attacks signals a shift from theoretical risk to immediate operational threat. The gap between the volume of AI-discovered vulnerabilities and the rate of upstream patching highlights a critical verification bottleneck that enterprises must manage through internal triage rather than relying solely on maintainers. Meanwhile, the aggressive capital deployment by DeepSeek and the launch of Mistral's sovereign-aligned model suggest that geopolitical competition is now being fought through infrastructure sovereignty and talent consolidation, not just model performance. For decision-makers, this implies a need to diversify model providers and invest in independent verification capabilities to mitigate both security and supply-chain risks.

This material was produced automatically by a large-language-model system from the public sources listed below; it is AI-generated content and may contain inaccuracies — verify facts against the original sources.

Sources