AI cybersecurity tools

Cybersecurity researchers at manifold.security warn that AI agent skills, MCP servers and plugins can conceal malicious components capable of hijacking automated workflows. The analysis outlines how compromised or unvetted extensions could lead agents to leak data or take harmful actions without user awareness, prompting renewed calls for vetting and supply-chain security in AI tooling.

A blog post lays out a strategy for conducting open-source investigations into cryptocurrency-related crime, combining OSINT techniques with AI tools. The item is tagged under digital investigations, cybersecurity, cybercrime and online investigations, suggesting it is aimed at researchers and analysts tracking illicit crypto activity. Discussion so far appears limited, with the post circulating mainly among the online investigations and cybersecurity community.

Cisco Talos researchers have identified 84 malware samples designed to manipulate large language model-based analysis systems. The samples use techniques ranging from deceptive embedded comments to exploitation of AI guardrails, aiming to fool automated security tools that rely on LLMs to inspect suspicious code. It is an early sign that attackers are adapting their methods specifically for AI-assisted cyber defence.

Bank regulators are increasing their oversight of how lenders use artificial intelligence, citing growing concerns that AI systems could expose banks to cyberattacks and operational failures. The tougher scrutiny reflects worries that rapid AI adoption in finance may outpace existing risk controls, and supervisors are expected to push banks for stronger safeguards and clearer governance of AI tools.

Anthropic is broadening access to its Claude AI system for vetted cybersecurity teams, in a move reported by The Hacker News. The announcement comes alongside figures from Glasswing, which says its use of the technology has uncovered 129,000 software flaws. The rollout signals Anthropic's effort to position Claude as a defensive security tool while keeping it away from malicious actors through screening of approved teams. Coverage is drawing attention from security professionals weighing AI's growing role in vulnerability discovery.

Microsoft has introduced a new interactive canvas for Excel Copilot, extending its AI assistant in Microsoft 365. The feature brings natural language report generation, guided skill building, and more advanced data cleaning tools directly inside spreadsheets. Early coverage highlights the canvas as a step toward making complex data analysis more conversational and accessible for everyday Excel users.

Anthropic has launched a free AI-powered vulnerability scanner for open-source projects. The tool uses AI to automatically find security flaws in open-source code, which security teams traditionally review manually. The announcement was reported by security outlets including The Hacker News, with attention focused on how AI firms are extending their models into cybersecurity tooling for under-resourced developer communities.

Posts are circulating about Google Gemini 4 Argon, described as a new AI model that reportedly matches OpenAI's GPT-6 Astra in performance while offering lower API pricing and integration with a tool called Antigravity. The claims are shared via a cybersecurity news site, but no official confirmation from Google of the model's specifications or pricing has been verified.

A new cybersecurity tool called Decepticon has been introduced as an autonomous red team agent that integrates Anthropic's Claude and OpenAI's Codex. The tool uses large language models to automate offensive security testing, running penetration-style tasks without constant human direction. Its release highlights the growing use of AI agents in security work, raising both interest in faster testing and concerns about offensive capabilities being widely available.

Cybersecurity firm CrowdStrike says a suspect believed to be a 26-year-old based in China used AI tools to carry out cyberattacks on South Korean banks. The report links the intrusions to a single individual who allegedly leveraged artificial intelligence to assist the hacking operations, highlighting growing concern over AI-enabled cybercrime and its implications for financial institutions and regional cybersecurity.

The Healthcare Sector Coordinating Council has published an AI clinical resilience playbook aimed at helping healthcare organizations deploy artificial intelligence safely. The guidance covers managing patient safety risks and cybersecurity threats that come with clinical AI systems. It is intended to give hospitals and other providers a practical framework for adopting AI while protecting patients and network infrastructure.

New Flashpoint research found that 65% of threat intelligence teams say alert noise is their biggest drag on productivity, while 50% expect their budgets to grow. The report argues the bottleneck is workflow rather than funding. It also found nearly half of teams are piloting or running agentic AI, with practitioners demanding proof that tools deliver real results before wider adoption.

Agentic AI — autonomous systems that can plan and execute tasks with minimal human oversight — is emerging as a major force in cybersecurity. Reporting highlights how these tools can detect threats, respond to incidents, and automate defense work faster than traditional software, while raising new questions about oversight and the risks if such systems are abused by attackers.

A commentator argues that organizations across the board lack visibility into the identities existing across their systems, a problem growing more acute as AI agents multiply. Sentinel, an open-source identity governance tool built for AI-agent environments, is being highlighted as a response, prompting discussion about identity sprawl and oversight in cybersecurity circles.

Cybersecurity coverage highlights concerns that AI-driven vulnerability patching lacks reliable ways to measure effectiveness. The argument is that grading methods for automated patch tools are poorly defined, making it hard to know whether AI-generated fixes actually resolve security flaws or introduce new risks. Security practitioners are weighing what better evaluation standards would look like as AI adoption in vulnerability management accelerates.

Intelligent CISO reports on the challenge of securing artificial intelligence in education, focusing on how schools and universities can protect students, their personal data and digital identities as AI tools spread through classrooms. The piece highlights growing pressure on education providers to manage cybersecurity risks while adopting new technology.

South Korean daily Chosun Ilbo reports that cybersecurity startups are seeing rapid growth as hackers increasingly use artificial intelligence to attack financial institutions. The report highlights rising demand for AI-driven defence tools as banks and fintech firms confront more sophisticated, automated attacks, prompting fresh investment and attention on the security startup sector.

A Chinese developer has closed the source code of its ARTEX AI agent following a hack that hit a South Korean bank. The move reverses any open availability of the tool and comes amid heightened scrutiny of AI software linked to cybersecurity incidents in the financial sector.

Law and technology librarian site BeSpacific has published its weekly cybersecurity roundup for October 10, 2026, compiled by Pete. The five highlights include a comparison of Mac and PC for privacy, Google's SynthID tool that detects AI-generated fakes for free, and news that the FBI uncovered the source of a compromise involving its jobs site.

WIRED reports that artificial intelligence is increasingly being used as a defensive tool to disrupt and deceive cybercriminals, turning hacking techniques back on the attackers. As AI-driven scams and attacks grow, security experts and companies are deploying AI systems that waste criminals' time, detect fraud, and mislead those attempting breaches, marking a shift toward AI-versus-AI conflict in cybersecurity.

Anthropic has released a free vulnerability scanner for open-source software. The tool is aimed at helping developers find security weaknesses in open-source projects at no cost. The announcement has drawn attention from the software and cybersecurity communities, with observers weighing what the offering means for the wider security tooling market.

Security researchers have shown that a straightforward 'Repeat-After-Me' prompt can bypass the safeguards of six widely used AI agent frameworks, allowing attackers to make the agents ignore their instructions and carry out unintended actions. The finding highlights how vulnerable agentic AI systems remain to prompt injection, and cybersecurity commentators are citing it as evidence that current defenses for autonomous AI tools are not ready for deployment in sensitive settings.

Chinese threat intelligence firm ThreatBook has acquired CyberStrikeAI, a platform that uses artificial intelligence to run automated penetration testing. The deal signals continued consolidation in the cybersecurity industry as vendors race to fold AI-driven offensive security tools into their portfolios. Details on the terms of the acquisition and integration plans have not been disclosed.

Software supply chain security firm Insignary has launched Clarity AIR, a new tool designed to identify undeclared open-source components and code generated by AI in software projects. The announcement has been picked up across multiple technology and cybersecurity outlets. The product targets a growing compliance concern: development teams increasingly use AI assistants and open-source libraries without documenting them, creating security and licensing risks for companies.

Anthropic has announced a new initiative focused on cybersecurity for critical infrastructure, which it is releasing as open source. The program, called the Cyber Mission, is aimed at helping protect essential systems from cyber threats. Further details about its scope, partners, and technical capabilities were not immediately available.