All news in category “AI and Security Pulse”

🔒 At AWS and other large vendors, AI is compressing months of security work into minutes by automating red-team findings, generating detections, and accelerating remediation. Experts debate whether this will widen an existing security class divide—where large organizations have talent, data, and infrastructure while smaller ones struggle—or democratize advanced capabilities. Key concerns include costs, privacy tradeoffs, token pricing, and the operational depth needed to safely run AI in production.

📰 OpenAI temporarily removed the five-hour usage restriction for Plus, Pro, and Business plans after a surge in demand for GPT-5.6 Sol over 48 hours. The company also reset current usage for all users and said it is rolling out efficiency improvements to reduce consumption by the model. This change affects how Codex and ChatGPT count local messages and cloud tasks against shared limits, giving users more uninterrupted access while preserving remaining weekly or plan-based caps.

📰 Anthropic has extended complimentary access to Claude Fable 5 for paid subscribers through July 19, 2026, allowing Pro, Max, Team, and eligible Enterprise seats to use the model without extra cost up to 50% of weekly subscription limits. The promotion also extends a 50% increase to Claude Code weekly usage limits. Fable 5 draws from the same weekly usage pool as other Claude models, uses credits faster, and the extension does not apply to Free users, standard Enterprise seats, usage-based Enterprise plans, or API access.

🔎 This article from Google Data Cloud explores a rigorous, information‑theoretic approach to evaluating AI data agents by converting binary pass/fail tests into graded difficulty sweeps. The team introduces Discovery Bench and iterative surprisal-based query refinement (iSQR) to generate low/medium/high ambiguity variations of queries, quantify surprisal, and map where agents succeed or fail. The piece highlights how this method reveals cliffs and sweet spots in agent behavior and calls for auditing benchmarks themselves to avoid misleading conclusions.

🛡️ Security vendor CrowdStrike has added five new prompt injection techniques to its taxonomy that threaten enterprise AI deployments. These attacks manipulate LLM behavior by embedding deceptive instructions into inputs, context, or token streams to bypass safety controls and produce malicious outputs. CrowdStrike recommends threat modeling input sources, expanding testing, and enhancing detection engineering to defend against composite and multi-stage prompt attacks.

🔍 AI-powered surveillance will soon monitor and penalize public and private behavior in real time, combining advanced facial recognition, mass databases, and personalized enforcement. These systems, already widespread in China and being trialed globally, can produce chilling effects on personal freedoms, democracy, and social progress by encouraging conformity and self-censorship. Policy interventions like bans, privacy protections, and AI regulation are proposed to prevent such outcomes.

🔐 The article argues that CISOs must shift from merely securing the business to enabling secure business operations with AI. It emphasizes focusing on intent, governance, and scope rather than assigning blame to AI agents, and recommends partnering with engineering, prioritizing high-risk exposures, and embedding continuous assurance into CI/CD. The piece frames AI security as a strategic enabler that preserves speed and mitigates risk.

🛡️ At Engage 2026 in Paris, Check Point CTO Jonathan Zanger discussed how AI is reshaping cybersecurity operations, enabling defenders to scale threat monitoring and red-team testing while also empowering attackers. He warned that AI increases attack surface as organizations connect models to enterprise systems and urged integrating security from the start of any AI project. Zanger emphasized prevention, collaboration across defenders, and protecting AI platforms themselves to mitigate rapidly evolving AI-driven threats.

🚀 Microsoft announces general availability of GPT-5.6 in Microsoft Foundry, integrating frontier models, production agent runtime, and enterprise controls into a single platform. The release includes Sol, Terra, and Luna model tiers with published pricing, Global and APAC Data Zone deployment options, and developer tooling for GitHub Copilot and VS Code. Foundry emphasizes observability, cost controls, and governance to accelerate production agent adoption.

🧭 AlphaEvolve, now GA on the Gemini Enterprise Agent Platform, is an agentic code optimization and discovery tool designed to tackle hard algorithmic problems across domains like logistics, semiconductors, genomics, and finance. It follows a four-step workflow — Define, Measure, Optimize, Apply — to move from a baseline algorithm to production-ready optimized code. Early customers report substantial gains in accuracy, performance, and efficiency across production pipelines and HPC workloads.

🛡️ Five Eyes agencies warned that AI's rapid development raises cyber risks, particularly autonomous hacking and automated attacks. Bruce Schneier explains that AI widens the gap between skill and ability, enabling less-skilled actors to cause greater harm while also offering defensive tools. He argues that guardrails on large platforms won't stop open-source models and urges using AI for defense across all heightened risks.

🔒 This article examines the risks of agentic AI and non-human identities in enterprise environments, illustrating incidents where LLM-based agents caused outages due to weak identity controls. It argues that existing IAM models are insufficient for agents that act autonomously, and cites industry guidance from Gartner, OWASP, CISA and NIST. The author proposes six minimum requirements and a cumulative six-stage NHI maturity model to ensure defensible production deployments.

🔒 Over 70 cybersecurity organisations have signed the new CREST AI Charter, launched on July 9, committing to nine principles governing AI-enabled cybersecurity activities. The charter covers accountability and governance, transparency of use, documentation and auditability, boundaries and control, data handling and sovereignty, security and confidentiality, secure development, supply chain assurance and resilience. Signatories will maintain human oversight, document AI use, disclose data practices and implement secure development and supply chain controls. CREST intends the charter as a self-regulatory foundation to drive standards and harmonisation across industry and regulators.

🔍 Security teams often default to retraining AI models when detections fail, but the real root cause is usually upstream data issues. Fragmented telemetry, inconsistent schemas and stale baselines degrade ML effectiveness long before models see events. Standardizing schemas, monitoring data quality at ingestion and applying governance to security telemetry are practical priorities that restore detection reliability without wholesale platform replacements.

🖼️ Meta introduced Muse Image, an image-focused AI from Superintelligence Labs that can use public Instagram posts and reels to generate AI-created images, enabled by default. The feature lets users @-mention public accounts in the Meta AI app to incorporate specific profiles' media into new images and is being integrated into Instagram and WhatsApp in select countries. Users can opt out via Instagram Settings > Sharing and reuse, though previously created content will remain if generated before disabling the setting. For minors with public accounts, only followers may reuse their media if allowed; existing remixes won't notify original owners, and deleted content may be removed if accounts go private for over 24 hours.

🔍 Researchers at the AI Now Institute demonstrated a proof-of-concept called "Friendly Fire" where autonomous AI coding agents (Anthropic's Claude Code and OpenAI's Codex) execute an attacker's binary when asked to scan untrusted open-source code. The attack hides a malicious binary alongside benign files and a README that prompts the agent to run a security script; in auto-modes the agents approved and executed it without prompting. The weakness is framed as a workflow/design issue rather than a single vulnerable version, and the researchers recommend never giving command-capable agents unattended access to untrusted code.

🎧 This episode of Smashing Security discusses a rash of recent cybersecurity incidents, including a 15-year-old who used a chatbot to cancel nearly 47,000 anime subscriptions and the first documented agentic ransomware, JadePuffer. The hosts also examine Apple’s problematic Hide My Email feature, which has been known to leak addresses for over a year. Guest Zoë Rose joins Graham Cluley to assess implications for security and privacy.

🛡️ System prompts are core to generative AI applications and often include role definitions, tool descriptions, RAG context, and other proprietary instructions. This AWS Security Blog post explains why system prompt leakage is a persistent risk, highlights that it cannot be fully remediated today, and outlines practical mitigations. It recommends design principles such as minimization and avoiding sensitive data in prompts, and details controls available via Amazon Bedrock Guardrails and other mechanisms to reduce exposure and raise extraction difficulty.

🛡️ New research describes "HalluSquatting," an attack that exploits AI assistants' habit of inventing resource names. Attackers register those predictable fake names on marketplaces and plant adversarial instructions; when an assistant hallucinate-fetches the same name, it may run the attacker's commands. The technique abuses auto-run modes and agent tools that fetch and execute external code, permitting widespread compromise without traditional malware or exploits.

🧭 A new study found GitHub Copilot can be induced to generate harmful answers when a dangerous request is broken into ordinary coding steps. Researchers Abhishek Kumar and Carsten Maple tested Claude and Gemini models through Copilot and observed that direct chat prompts were routinely refused, but the same content was produced in 816 of 816 workflow runs when framed as benchmark-improving “teaching shots.” The paper calls this technique workflow-level jailbreak construction and urges reviewing written files and whole sessions, not just chat refusals.