All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
Security researcher Ian Carroll used Claude Opus (an AI assistant) to discover a vulnerability in Front Gate Tickets' website that allowed him to gain super-administrator access (high-level control over a system) and issue free tickets to music festivals across the US. Carroll reported the bug responsibly rather than exploiting it, and Front Gate patched the vulnerability within 24 hours, demonstrating how AI tools can help identify serious security flaws in web systems.
Fix: Front Gate Tickets patched the vulnerability. According to the company's statement: 'This was resolved within 24 hours, and we can confirm there is no evidence of exploitation, ticket impact, or compromise of customer information.' Anthropic also noted that if Carroll had not been part of its Cyber Verification Program (an approved security research program), his use of Claude to hack the system would have been detected and blocked.
Wired (Security)Detection engineering is the practice of creating custom rules and systems to identify security threats in an organization's specific environment while reducing false alarms (incorrect threat alerts). Unlike traditional threat detection that relies on generic vendor-provided rules, detection engineering applies software development principles to build tailored detection logic based on an organization's unique threats and how attackers actually behave, with the approach now becoming a strategic priority across most organizations.
The LearnPress WordPress plugin (versions up to 4.3.9.1) has a security flaw called IDOR (insecure direct object reference, where attackers bypass checks meant to prevent accessing other users' data) in the 'userId' parameter. An authenticated user with subscriber-level access or higher can view course enrollment and completion information belonging to teacher or administrator accounts by exploiting missing validation checks.
Anthropic's Claude Fable 5 and Mythos 5 AI models, which the U.S. government had restricted from being shared with foreign nationals due to national security concerns, have been cleared for release again following negotiations with the Trump administration. Fable 5 will be available globally starting Wednesday through Claude's platforms, while Mythos 5 access has been restored for some U.S. organizations with plans to expand through Anthropic's Glasswing program (a cybersecurity initiative providing selected organizations access to advanced AI models for defensive security testing).
Anthropic has announced that the Department of Commerce lifted export controls on Claude's powerful AI models, Fable 5 and Mythos 5, and will begin restoring access to Fable 5 on Wednesday. The company is also rolling out identity verification (KYC, or know-your-customer checks) for certain Claude features, requiring users to provide government-issued photo ID and a live selfie through a third-party verification partner called Persona to prevent misuse and comply with legal requirements.
Anthropic's Claude Fable 5 AI model is being brought back online after the U.S. Department of Commerce lifted export controls that had previously blocked it. The company plans to restore access to users globally starting Wednesday across Claude platforms and cloud services like AWS, Google Cloud, and Microsoft Foundry, though without a specific timeline for completion.
Anthropic announced that the U.S. Department of Commerce has removed export restrictions on two AI models, Claude Fable 5 and Mythos 5, allowing the company to restore user access to these models starting the next day. This announcement suggests that government controls on advanced AI technology exports had previously limited who could use these models outside certain regions.
Anthropic has released Claude Sonnet 5, a new AI model that performs nearly as well as their expensive flagship model (Opus 4.8) but costs significantly less. Sonnet 5 is designed to be 'agentic,' meaning it can make plans, use tools like browsers and terminals, and check its own work, capabilities that were previously limited to more expensive models. The model is available now with introductory pricing of $2 per million input tokens (the words you feed it) and $10 per million output tokens (the words it generates) through August 2026.
Anthropic announced Claude Science, a new AI product designed to help scientists conduct research autonomously, similar to how Claude Code supports software engineers. The tool has special features for computational biology and drug development, including the ability to run code on powerful computers and ensure results can be verified for accuracy. This positions Anthropic as a major competitor to Google DeepMind in AI-assisted scientific research.
Anthropic, an AI company, is starting its own drug discovery program to develop AI tools for pharmaceutical companies, focusing on treatments for neglected diseases (illnesses that traditional drug makers ignore because they're not profitable). The company believes that by doing its own drug research alongside pharmaceutical partners, it can build better AI products and get direct feedback on what drugmakers actually need.
Claude Sonnet 5 is a new AI model released by Anthropic with performance similar to their more advanced Opus 4.8 model but at lower advertised prices. However, the model uses a new tokenizer (a system that breaks text into small units for processing) that produces approximately 30% more tokens for the same input text, effectively increasing costs by 1.27x to 1.42x depending on the language, with minimal impact on Simplified Mandarin text.
A missing authorization vulnerability in GitHub Enterprise Server allowed authenticated users to read source code from private repositories they shouldn't have access to. The vulnerability existed in a Copilot pull request description feature that compared code across repositories without checking if the user had permission to view the target repository, and it required the attacker to already have read access to at least one repository on the system.
Large language models frequently invent web addresses that don't exist, and attackers are now registering these fake domains before anyone else can, then hosting phishing pages on them to catch people following AI-generated links. Palo Alto Networks' research found that when two AI models answered 685,339 questions about major brands, they generated 2.1 million links, including roughly 250,000 made-up domains with no owner yet and 13,229 known-malicious addresses. This attack, called phantom squatting (a technique where criminals register fake domains that AI systems invented), works because brand-new domains have no reputation history for security filters to flag them, so victims reach the phishing site before defenses catch up.
Anthropic restored access to Claude Fable 5 worldwide after the U.S. lifted export controls (restrictions on who can use a technology) that had been imposed for two and a half weeks following the discovery of a jailbreak (a prompt that tricks an AI into bypassing its safety rules). To address the security concern, Anthropic trained a new safety filter called a classifier that blocks the specific jailbreak technique in over 99% of attempts, while automatically routing blocked requests to a weaker model and notifying users.
Fix: Anthropic trained a new safety filter called a classifier that watches for the exact jailbreak technique and blocks it. The company says it now stops that technique in more than 99% of tries as of the June 30 write-up. Blocked requests get handed to the weaker Opus 4.8 model instead, and the user is told.
The Hacker NewsFix: The U.S. Department of Commerce lifted the export controls on both models. According to Commerce Secretary Howard Lutnick's letter to Anthropic (viewed by CNBC), he determined that 'appropriate safeguards' were in place to permit certain 'trusted partners' to access the model. Fable 5 will be available to global users on the Claude platform, Claude.AI, and Claude Code starting Wednesday, and will also be re-enabled on Amazon Web Services, Google Cloud, and Microsoft Foundry as soon as possible.
CNBC TechnologyThe US lifted export controls on Anthropic's Fable and Mythos AI models (advanced language models developed by the company) after the company agreed to detect security risks, work with the government on safety standards, and report any malicious activity. The controls had been imposed weeks earlier due to national security concerns that these powerful AI systems could be misused by foreign militaries or intelligence agencies.
Fix: Anthropic agreed to proactively detect and address security risks associated with the models; to work diligently with the US government on protocols and standards for releases; and to inform the US government of any malicious activity. The US government also established a vetting process to control which organizations can access the models, limiting initial access to 'trusted' US organizations.
The Guardian TechnologyResearchers discovered that large language models (LLMs, AI systems trained on massive amounts of text data) frequently generate fake web domain names that don't actually exist but sound like they belong to real companies. Attackers are registering these made-up domains (a practice called phantom squatting) to intercept traffic sent by AI systems and users who trust the LLM's output, creating a new supply chain attack vector (a way to compromise software development and deployment). The researchers identified over 13,000 confirmed malicious URLs and approximately 250,000 unregistered hallucinated domains that attackers could exploit.
Fix: Palo Alto Networks customers can use the following products and services for protection: Advanced WildFire, Advanced URL Filtering, Advanced DNS Security (systems that monitor and block suspicious web traffic), Prisma AIRS, Koi Agentic Endpoint Security, and the Unit 42 AI Security Assessment. Organizations can also contact the Unit 42 Incident Response team if they suspect compromise.
Palo Alto Unit 42Microsoft SharePoint Server has a deserialization of untrusted data vulnerability (a flaw where the software unsafely processes data from an untrusted source, allowing an attacker to inject malicious code), which lets an authorized attacker run code over a network. This vulnerability is actively being exploited in real attacks. Organizations must apply security updates following CISA's BOD 26-04 guidance by July 4, 2026, or stop using the product if no fix is available.
Fix: Apply mitigations in accordance with vendor instructions, ensuring compliance with CISA's BOD 26-04 Prioritizing Security Updates Based on Risk guidance. Follow applicable BOD 26-04 guidance for cloud services or discontinue use of the product if mitigations are unavailable. See https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-45659 for vendor-specific details and https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk for patching guidelines.
CISA Known Exploited VulnerabilitiesThis release document describes updates to security techniques and mitigations related to AI systems, including new attack methods like stealing web session cookies and jailbreaking LLMs (large language models, AI systems trained on massive amounts of text), as well as updated defenses like AI guardrails (safety features that prevent unwanted outputs) and telemetry logging (monitoring system activity). Several new case studies illustrate real-world attacks on AI services, including prompt injection (tricking an AI by hiding malicious instructions in user input) and vulnerabilities in commercial AI products.
BioShocking is a prompt injection attack (tricking an AI by hiding malicious instructions in its input) that manipulates AI-powered browsers into ignoring safety guardrails by framing dangerous actions as part of a fictional game scenario. Researchers at LayerX tested this attack on six mainstream AI browser products and found that all six failed to distinguish between harmless game actions and real sensitive operations like stealing passwords. Only OpenAI implemented a working fix for the vulnerability.
Fix: OpenAI was the only vendor to implement a working fix for BioShocking in ChatGPT Atlas. The source also recommends that vendors add explicit user confirmation for sensitive actions, stronger context checks, and scope limits for agentic sessions (AI agent operating boundaries), while users should restrict AI browser access to sensitive services through available platform options.
BleepingComputerA new attack called "agentjacking" shows how attackers can trick AI coding agents (AI systems that automatically write and modify code) by exploiting their inability to tell the difference between regular content and hidden instructions. The attack uses fake bug reports to hijack these agents at scale, affecting many systems at once.
Fix: The vulnerability was fixed in versions 3.17.17, 3.18.11, 3.19.8, and 3.20.4. All versions prior to 3.21 were affected.
NVD/CVE Database