Anthropic reveals its invisible Claude text watermark mechanism, utilizing a probability-based algorithm to comply with the EU AI Act without altering meaning.

Related Posts:

• Anthropic Revenue Surges 14x to $11.5B in Q2 2026 Ahead of IPO
• Stripe Finalizes $7 Billion Acquisition of OpenRouter
• SpaceX Acquires AI Coding Startup Cursor

The post Anthropic Claude Text Watermark: An Invisible Statistical Algorithm appeared first on Daily CyberSecurity.

A June 2026 research review found that AI chatbot warning labels may be a weak safeguard for organization-backed AI advisors, raising new audit questions for IT, security, and compliance teams.

The post AI Chatbot Warnings May Not Stop Hallucinations, Researchers Say appeared first on TechRepublic.

By Niall Browne, CEO and Founder, AIBound

Shadow AI is accelerating alongside artificial intelligence (AI) adoption at a pace that has outgrown most enterprise governance models. Artificial intelligence (AI) adoption is accelerating at a pace that has outgrown most enterprise governance models. According to the World Economic Forum, 87% of organizations report that AI-related vulnerabilities are now the fastest-growing cyber risk. Part of this surfaces with  employees increasingly deploying autonomous AI agents that connect to MCP servers and external AI that security teams have never assessed, quietly piping sensitive corporate data into systems no one in IT has ever audited — and no one in the C-suite knows exist. This increase in Shadow AI is creating systemic enterprise risk that can lead to unforeseen costs. Compliance frameworks like the Artificial Intelligence Act of the European Union (EU AI Act) take full effect this year introducing penalties up to 7% of global annual revenue for unmanaged AI. As regulatory frameworks begin to align with the realities of increased AI adoption, enterprises need to account for decentralized AI usage that operates outside traditional controls. This requires software that allows greater visibility, organization, and control into how AI is used and tracked across environments.

Shadow AI Is Creating a New Enterprise Attack Surface

The traditional security stack was built for a world that no longer exists — one with known assets, centralized systems, and software that asked permission before it ran. As new tools are introduced independently, usage levels evolve quickly without system checks or visibility into how these tools interact with sensitive data. Research indicates that 75% of CISOs have discovered unsanctioned GenAI tools in their environments, and only 5% feel confident they could contain compromised AI agents. Because of how easy these platforms are to access and require little onboarding, adoption is happening across teams at a rapid rate without IT involvement. Other security issues lie with employees integrating workflows with personal AI agents. These deployments allow sensitive information to be leaked or directly inputted into agents without security knowledge. Without a system in place for organizations to continuously track and evaluate how AI is being used across their enterprise systems, CISOs are left without visibility of their attack surfaces. The result is a slow-motion breach: data leaking, compliance crumbling, and governance reduced to a slide deck nobody enforces. Recurring data leaks and breaches via AI reveal the need for solutions that address this gap. Popular AI agents like ChatGPT for example, revealed a ‘ShadowLeak’ vulnerability that allowed sensitive email data to be breached through a zero-click attack. Other short lived features that rolled out last year allowed conversation sharing, leaving employee info, internal corporate strategies, and other sensitive data to be shared and indexed by search engines. Although this option only was available for a day, it was estimated that over 100,000 private chats were affected and able to be viewed with a simple search, allowing any sensitive information inputted to be publicly accessible. Other recent breaches include a Microsoft 365 Copilot bug allowing AI assistants to summarize emails labeled confidential, bypassing data loss prevention policies set up by organizations. Microsoft confirmed that a code issue allowed confidential emails data to be accessed despite organizational securities put in place. These agents are live and operational with local access to files, systems, commands, and APIs capable of executing tasks and retrieving data without clear oversight control. As AI usage continues to expand at accelerating rates, organizations need a way to better understand how these tools are used across their environments. No CISO has ever defended a perimeter they couldn't see. Shadow AI is the new perimeter — and most security teams are flying blind. Without a comprehensive inventory and control of AI usage, security teams are unable to accurately assess risks and enforce policy to maintain compliance.

Shadow AI Demands Continuous Visibility and Independent AI Control Planes

This is where adoption of independent AI Control Planes becomes vital. Independent AI Control Planes provides a way to continuously identify and assess AI activity giving security teams the visibility needed to manage emerging risks. It enables organization and categorization of AI usage across enterprises without relying on the manual entry and tracking that existing platforms demand — work no security team in a fast-moving environment can realistically keep up with. It’s undeniable: Shadow AI is not a future problem — it is already running inside your enterprise, on assets you don't own, through agents you never approved, touching data you are responsible for protecting. Every day without continuous, autonomous AI discovery is a day your attack surface grows faster than your governance can chase it. Regulators won't wait. Attackers already aren't. The CISOs who win the next 24 months will be the ones who stop pretending policy equals control and start operating on a simple truth: if you can't see it, you can't secure it — and right now, most of AI is invisible.

Disclaimer: The views and opinions expressed in this guest article are solely those of the author and do not necessarily reflect the official policy or position of The Cyber Express. The information shared is intended for industry discussion and awareness purposes only.

The European Union has reached a provisional agreement to amend parts of the EU AI Act, introducing simplification measures for businesses while also expanding restrictions on harmful AI applications, including so-called “nudifier” apps and AI-generated child sexual abuse material. The agreement, reached early Thursday by negotiators from the European Parliament and the Council, forms part of the EU’s broader “digital omnibus” package aimed at refining the implementation of the bloc’s landmark AI legislation. The updated proposal seeks to reduce compliance burdens and legal uncertainty for AI providers while maintaining the AI Act’s core risk-based framework. Lawmakers said the changes are designed to make the rules more practical without weakening safeguards tied to safety, privacy, and fundamental rights.

EU AI Act Deadlines Pushed to Reduce Legal Uncertainty

One of the biggest changes under the proposed amendments is the postponement of several obligations linked to high-risk AI systems. Under the revised timeline, rules for AI systems classified as high-risk due to their use cases will now apply from 2 December 2027. These systems include AI deployed in biometric identification, critical infrastructure, education, employment, law enforcement, and border management. Meanwhile, AI systems used as safety components under sector-specific EU product safety laws will face compliance obligations from 2 August 2028. The agreement also delays watermarking obligations for AI-generated content until 2 December 2026. The European Commission had earlier proposed a February 2027 implementation date. Watermarking tools are intended to help identify and trace AI-generated images, audio, and video content. Lawmakers said the postponements are necessary to ensure technical standards and implementation guidance are fully in place before the rules become enforceable.

EU Bans Nudifier Apps and AI-Generated Abuse Content

A major part of the agreement focuses on tightening restrictions around harmful AI-generated sexual content. Negotiators agreed to ban AI systems designed to create child sexual abuse material or generate explicit deepfake content involving identifiable individuals without consent. The restriction covers images, video, and audio content. The EU AI Act ban specifically applies to companies placing such AI systems on the EU market, providers failing to include reasonable safeguards against misuse, and users deploying the systems to create illegal or non-consensual explicit material. The decision directly targets “nudifier” apps, which use AI to digitally remove clothing or generate fake explicit imagery of individuals. Companies operating such systems will have until 2 December 2026 to comply with the new requirements. Michael McNamara, co-rapporteur for the Civil Liberties, Justice and Home Affairs committee, said the agreement strengthens the EU’s ability to act against AI systems that threaten human dignity and fundamental rights. “I’m pleased that this morning we reached an agreement on the AI Omnibus,” McNamara said. “Alongside simplification measures, we are banning nudification apps, a key part of the Parliament’s mandate, and, of course, the creation of child sexual abuse material using AI systems.”

Simplification Measures for AI Providers and SMEs

The amendments also introduce several simplification measures intended to reduce overlapping compliance requirements for companies developing AI technologies. Under the new framework, machinery products with AI features will no longer need to comply separately with both the EU AI Act and sector-specific safety laws if existing safety rules already provide equivalent protection. Lawmakers also narrowed the definition of “safety component” within the EU AI Act. This means AI functions designed only to assist users or improve product performance will not automatically be classified as high-risk unless their failure creates health or safety risks. Another change allows companies to process personal data where strictly necessary to detect and correct bias in AI systems, provided appropriate safeguards are in place. The agreement further extends certain exemptions previously available only to small and medium-sized enterprises (SMEs) to small mid-cap companies. EU officials said the move is intended to help startups and growing technology firms scale AI innovation more easily within Europe. Arba Kokalari, co-rapporteur for the Internal Market and Consumer Protection committee, said the revised rules strike a balance between innovation and regulation. “With this agreement, we show that politics can move just as quickly as technology,” Kokalari said. “We now make the AI rules more workable in practice, remove overlaps and pause the high-risk requirements.”

Next Steps for the EU AI Act Amendments

The provisional agreement still requires formal approval from both the European Parliament and the Council before it can become law. EU lawmakers are aiming to finalize adoption before 2 August 2026, which marks the scheduled start date for existing high-risk AI system rules under the original AI Act framework. The negotiations are part of the EU’s continuing effort to shape global standards around artificial intelligence governance while addressing concerns related to safety, transparency, and misuse of generative AI technologies.