The UK’s @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6 Sol. The models attempted to complete an assignment in a setup where their normal safeguards were removed and they were deliberately given internet access. AISI reports that the models “engaged in sustained, potentially harmful activity directed at real people and organisations”. We’re grateful to AISI for their leadership in the important discussion about how to evaluate increasingly capable AI agents. We’re working closely with them to gather more details of the incident as we conduct our own investigation. Gaining a clear picture of Claude’s understanding of its situation—by examining its reasoning transcripts and running our own analyses—will help us identify the causes of its behavior. The prompts in the evaluation did not impose any specific restrictions on how the internet should be used. This and the removal of safeguards meant that the models were tested under “deliberately permissive conditions” that are not representative of any of our production models. Note that there was no evidence here of an escape from a secure environment. AISI’s disclosure of the incident can be found here:

🚨 Agent security alerts often drown in noise. 🧭 AgentLoop Audit separates real risks from noise: ꔷ Full session & tool facts ꔷ Contextual risk analysis ꔷ One-click evidence location ꔷ Impact traced by secret, app & user 🔗  #AIAgent# #Security# #AgentSecurity# #AISecurity# #LLMOps#

🔥From Black Hat Asia 2017 as a Speaker to Black Hat USA 2026 at Arsenal. Nine years later, still building, still breaking, still learning. #BlackHatUSA# #Arsenal# #AISecurity# #Web3Security#

AI Agents Agreeing With You Is Dangerous What happens when an AI agent stops challenging your assumptions and starts reinforcing them? One of the less obvious risks of AI adoption is overconfidence. When AI systems consistently validate what users want to hear, they can reinforce flawed decisions instead of helping people identify problems. Craig McLuckie's perspective on engineering discipline highlights why organizations cannot confuse agreeable AI behavior with reliable AI behavior. As agents become more deeply integrated into business processes, trust must come from verification, context, and enforceable safeguards. Topics covered: • AI sycophancy and overconfidence • The danger of unchallenged AI recommendations • Context optimization and reliable AI outputs • Building safeguards into enterprise AI systems Guest: Craig McLuckie — CEO & Founder, Stacklok Host: Brandon Krieger — CEO & vCISO Advisor 📺 Watch the full episode: 🎧 Listen: #ArtificialIntelligence#, #AIAgents#, #AITrust#, #AISecurity#, #Stacklok#, #CraigMcLuckie#

TL;DR: Instead of hunting vulnerabilities one by one, shrink the code that can be reached. Claude analyzed 25 open-source projects, reported 493 issues, and cut about 16% of OpenSSH's pre-authentication code. Title: Using Claude to reduce the attack surface of code URL: Key points ・🗺️ 25 projects and 5M+ lines, with 17,000+ config options inventoried (3,842 security-relevant) ・🔓 93 unnecessary pre-auth surfaces in default configs, plus 61 combos that silently weaken a pre-auth control ・⚙️ 162 option combinations accepted with no warning, 94 of them silently disabling a security control ・🌐 Hostname-based rules relying on reverse DNS fail open when it's off (4 projects) ・⏳ A config conflict dropped a certificate requirement for 21 years, undocumented ・🔐 OpenSSH: about 4,500 lines of experimental ML-DSA-44 code reachable by unauthenticated clients removed, shipped in 10.6 Making unknown bugs unreachable seems more practical than fixing them one at a time. #AISecurity# #OpenSSH#

Think you can break an AI application? Prove you can defend it too. Join us with @wiz_io on Sept 18 in Amsterdam for a hands-on AI security CTF. $10K total in Token Factory credits up for grabs 🏆 📍 Amsterdam 🗓️ Sept 18, 1:00 PM Register: #AISecurity# #CTF#

JSEF v1.10.0 is out 🎉 A model-agnostic LLM eval toolchain + 20+ new hard vulnerability samples. ✅ Swap any OpenAI/Anthropic-compatible model (GLM-5.3, Mythos, GPT…) ✅ One-click cross-model leaderboard (Recall/Precision/F1/accuracy) ✅ DeepSwe-style trials stability (Pass@1, N-run) ✅ Multi-state gate, JWT 4-step chain, async taint chain Star → #LLMSecurity# #AppSec# #WebSecurity# #JavaSecurity# #VulnerabilityDetection# #AISecurity#

Would you give an AI agent permission to book, cancel, or "optimize" real-world actions on your behalf today? 🗳️ Yes – with set limits 🗳️ Yes – full autonomy 🗳️ No way 🗳️ Only for work For people who voted "No": Web2 integrations require sharing master API keys with remote LLMs. One prompt injection or hallucination, and your account is wiped. How #SVPChain# makes "Yes" safe: You never hand over master keys. Agents get granular session tokens with hard spending rates and action whitelists signed locally via ⁠svpchain-mcp⁠. True autonomy doesn't need blind trust—verified onchain with SVPChain. #AIAgents# #AgenticAI# #Web3# #AISecurity# #OnChain# #SVPChain#

AI agents do more than browse. In cyber evals, agents took unsanctioned action on the live internet under permissive test settings. Web3 needs the lesson now: hard boundaries, monitoring, and human-readable approval moments. 🤖🛡️ #HootArk# #AISecurity# #Web3Security#

Can your organization clearly say how much it's spending on AI? For many, the answer is no. AI token usage is creating a new challenge for Finance and IT. Costs are harder to track, governance is fragmented, and visibility is limited. Join 1Password CFO Greg Henry and Global Advisory CISO Dave Lewis on July 21 to learn what leading organizations are doing to get AI consumption and spend under control. 🗓️ July 21 at 11:30 AM PT / 2:30 PM ET Register: #AI# #AISecurity# #SaaSManagement# #Finance# #1Password#