Loading story…
Organizations are deploying generative and agentic AI faster than governance capacity is expanding, leaving CISOs responsible for risks spanning sensitive-data exposure, shadow AI, compromised cloud accounts, insider activity, and SaaS collaboration platforms. Proofpoint reports that AI-governance responsibilities are growing without matching specialist resources, while a WalkMe survey found 45% of employees used unauthorized AI tools in the prior 30 days and 36% shared confidential data with them. Enterprises are responding with defined agent boundaries, continuous monitoring, logged human overrides, rapid circuit breakers, approved AI alternatives, and centralized internal LLM platforms.
Regulatory and assurance demands are making those controls more auditable. ISO/IEC 42001 provides a management-system model for organizational AI governance, and OWASP’s new GenAI Crosswalk maps 51 AI risks to controls across 25 frameworks and regulations. Under Article 73 of the EU AI Act, providers of high-risk systems may need to report serious AI incidents—including harmful downstream decisions without a cyber breach—within as little as two days for widespread incidents. Research also finds that technical safeguards alone remain insufficient: inference-time controls are not adequate against state-level adversaries, enterprise models often fail to refuse disguised prohibited requests, and shared AI-model foundations can undermine independent software assurance. Organizations are therefore emphasizing versioned specifications, independent validation, policy testing, traceable decision records, and human authority to halt systems.
Track how attackers are adapting to this technology.
15 events from the most recent confirmed update back to the earliest known activity.
OpenAI published a policy statement advocating mandatory, capability-based U.S. federal regulation for frontier AI, including independent assessments, cybersecurity protections, incident reporting, and national preparedness. It also said it had introduced full-trajectory monitoring and a mandatory alignment-evaluation gate before broader internal deployment for Astra.
Samar Ansari submitted a paper assessing 20 inference-time AI governance mechanisms across monitoring, verification, and enforcement. The assessment found 15 mechanisms had production technical foundations but none was adequate against a high-capability state-level deployer.
The second revision of Ghanem's paper on correlated failures from shared generative-model substrates was dated September 8, 2026.
The OWASP GenAI Security Project published its Crosswalk, mapping 51 generative-AI risks from four OWASP lists to controls across 25 standards, frameworks, and regulations. OWASP said the mapping tool is not a certification and requires use-case-specific validation.
EU AI Act Article 50 content-marking obligations and Article 73 serious-incident reporting obligations took effect. Article 73 requires high-risk AI providers to report serious incidents to national market-surveillance authorities, including certain harmful downstream AI outcomes without a cyber compromise.
Mohamed Chahine Ghanem submitted a paper proposing measurable independence and bounded autonomy for generative models used to build, defend, monitor, and test software.
Chase launched LLM Suite, an internal agentic platform for employees to query information, review documents, and draft specifications through a secured pipeline with AI controls.
Google deployed SynthID-Text watermarking in its Gemini models, according to the cited analysis.
Tim Pfaelzer of Veeam cited an example in which a hallucinating AI agent allegedly deleted an online retailer's backups and production data within nine seconds. The reported incident caused approximately 10 million lost orders and a 72-hour outage.
Researchers evaluated an open-source SynthID-Text implementation on two open-weight models and reported no measurable prose-quality effect beyond sampling variation, mixed code-quality effects, and near-chance watermark detection. They concluded that deployed vendor watermarking claims cannot presently be independently verified because testing access and governance mechanisms are unavailable.
A within-subject study of human reviewers assessing two AI-generated banking-service implementations found that approved specification, HLD, and LLD baselines did not materially change defect recall but increased attribution of findings to approved requirements. The governed baseline also significantly increased review time.
Proofpoint published its 2026 Voice of the CISO report, finding that 78% of surveyed CISOs consider GenAI a security risk and 79% are expected to manage AI risk without proportional increases in resources or expertise.
Meta described a compliance-domain AI-agent architecture that stores institutional knowledge in version-controlled files, uses explicit reasoning recipes and human checkpoints, and turns reviewed expert corrections into regression-tested updates without model retraining.
AIM Intelligence and academic and industry collaborators evaluated seven LLMs with 5,920 policy-alignment queries across eight fictional industries. The models generally answered allowed requests well but poorly refused adversarially disguised prohibited requests; policy-aware LoRA fine-tuning was the strongest evaluated mitigation.
Anthropic disclosed that Claude models released after the Article 50 effective date embed SynthID-Text-based watermarks in generated text by default and without a user opt-out.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
15 references tracked. Mallory keeps watching after this page renders.
itpro.com
Open sourceinfoq.com
Open sourcehelpnetsecurity.com
Open sourcearxiv.org
Open sourcecio.com
Open sourcesecuritymagazine.com
Open sourceaim-intelligence.com
Open sourceiso.org
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Organizations are rapidly integrating AI technologies into their operations, with studies showing a significant increase in AI adoption and the implementation of AI acceptable use policies. However, despite these advances, many companies struggle to effectively classify and protect data, and governance frameworks often lag behind the pace of AI deployment. Reports highlight that while 80% of organizations have established AI use policies, only a third feel confident in their data protection measures, and over half believe their data is not yet ready for AI. This gap between adoption and governance is further exacerbated by the acceleration of data growth driven by AI, with a notable rise in organizations managing petabyte-scale datasets. The lack of robust governance and holistic data management frameworks has led to increased risks, including the emergence of shadow identities and unmitigated security threats associated with AI tools. Experts emphasize the need for organizations to move beyond initial policy creation and embed comprehensive AI governance and dynamic data protection into their core operations. Without these measures, the benefits of AI could be undermined by vulnerabilities and operational blind spots, making it critical for security teams to proactively address these challenges as AI becomes ubiquitous in enterprise environments.
Jun 29, 2026
A global shift is underway as artificial intelligence becomes the top investment priority for cybersecurity among organizations, according to a recent PwC survey of nearly 4,000 business and technology executives. Sixty percent of respondents indicated that AI technologies are their primary focus for cybersecurity spending in the coming year, with key use cases including threat hunting, AI agents, event detection, and behavioral analysis. This surge in AI adoption is driven in part by ongoing skills shortages, with over half of organizations prioritizing AI and machine learning tools to close capability gaps. Additionally, 38% of companies are turning to managed service providers to access the necessary AI expertise. However, the rapid integration of AI into enterprise environments is not without significant risks. New research from LayerX reveals that AI has already become the leading uncontrolled channel for corporate data exfiltration, surpassing traditional risks like shadow SaaS and unmanaged file sharing. The report highlights that 45% of enterprise employees use generative AI tools, such as ChatGPT, Claude, and Copilot, with AI accounting for 11% of all enterprise application activity. Alarmingly, 67% of AI usage occurs through unmanaged personal accounts, leaving security teams with little visibility or control over sensitive data flows. The research found that 40% of files uploaded to generative AI platforms contain personally identifiable information (PII) or payment card information (PCI), and a significant portion of these uploads are conducted via personal accounts. Traditional data loss prevention (DLP) tools are often ineffective in this context, as they are not designed to monitor or control the new channels introduced by AI tools. Experts warn that the lack of governance and oversight around AI usage in enterprises creates substantial risks, including the potential for AI agents to access sensitive data beyond their intended scope if not properly managed. The PwC survey also notes that organizations are balancing investments between proactive technologies, such as monitoring and testing, and reactive measures like incident response and recovery. As AI becomes more deeply embedded in cybersecurity operations, the need for robust governance, least-privilege access for AI agents, and updated security controls is increasingly urgent. The evolving geopolitical landscape further complicates the risk environment, making it critical for organizations to understand both the opportunities and threats posed by AI in cybersecurity. The convergence of rapid AI adoption and insufficient controls underscores the importance of immediate action to secure enterprise data and workflows against emerging AI-driven threats.
Jun 29, 2026
Organizations are rapidly expanding the use of **AI agents**—systems that can execute multi-step tasks with limited human supervision—while governance, safety, and oversight controls lag behind. Deloitte’s *State of AI in the Enterprise* survey of 3,200+ business leaders across 24 countries reported **23%** of companies already using AI agents “at least moderately,” projected to rise to **74%** within two years, while only about **21%** said they have robust safety and oversight mechanisms in place. Separately, commentary warning about AI-enabled intrusion acceleration cited a purported “**GTG-1002**” campaign in which AI agents allegedly automated most of the intrusion lifecycle and compressed response windows, arguing that traditional SOC processes struggle against autonomous, high-velocity adversary tradecraft. Multiple other items in the set focus on broader *responsible AI* and policy concerns rather than a single security incident: an interview-style piece describes how “responsible AI” functions inside a large vendor’s product process, and another report highlights expert concerns about deploying LLM tools in **law enforcement** workflows (e.g., summarizing body camera transcripts or generating crime scene photo descriptions) given risks like hallucinations and bias. A separate business-leadership article frames cybersecurity and AI as strategic imperatives amid geopolitical instability but does not provide incident-specific or vulnerability-specific details. Overall, the material is best characterized as **governance and risk posture** coverage around agentic AI rather than a unified, verifiable breach or vulnerability disclosure.
Jun 29, 2026