Russian military intelligence unit APT28 used DNS hijacking to compromise thousands of SOHO routers, specifically targeting legacy TP-Link models across 23 US states. This operation highlights the critical security risk of using end-of-life hardware that no longer receives firmware updates to protect network traffic.
Zhipu AI has released GLM-5.2, an open-weight model that researchers claim matches Mythos in specific bug-finding and cybersecurity tasks. Because the model is open-weight and runnable on common hardware, it reduces the capability gap between US and Chinese AI while increasing risks of unsupervised misuse.
Federal agents located David Streever at a New York City hotel shortly after he entered the U.S. from Finland, despite him not returning to his primary residence. The incident raises questions about the specific surveillance tools and financial data access used by law enforcement to track individuals in real-time.
Bitcoin holders may receive airdrops from two upcoming fork attempts, including the eCash hard fork scheduled for August 21 and the BIP 110 soft fork proposal. Users should be cautious of potential replay attacks and security risks when importing private keys into unfamiliar software to claim these new assets.
Flock Safety has deployed over 100,000 AI-powered cameras nationwide that use natural language searching to track vehicles and individuals beyond simple license plate recognition. The system faces scrutiny over critical security vulnerabilities, lack of oversight in law enforcement access, and AI inaccuracies leading to wrongful police encounters.
Anthropic's Claude Fable 5 model is expected to return to public availability this week after the U.S. government previously restricted it via an export control directive. This reversal follows the restoration of the Mythos 5 model and signals a resolution between the Trump administration and Anthropic regarding AI safety safeguards.
Sakana AI has launched Fugu, an agent-orchestration model, while Chinese firm 360 unveiled Tulongfeng and Yitianzhen for cybersecurity. These releases capitalize on a U.S. export ban preventing non-Americans from accessing Anthropic's Mythos and Fable 5 models, signaling a shift toward local AI alternatives in Asia.
OpenAI is previewing GPT-5.6 to select partners in three variants: Sol for high-reasoning and cybersecurity, Terra for general use, and Luna as a low-cost option. The release follows a voluntary government review process and emphasizes aggressive jailbreak prevention and tiered pricing to lower the cost of high-performance inference.
The US government has authorized Anthropic to redeploy its Mythos 5 cybersecurity AI model to over 100 trusted organizations defending critical infrastructure. This move follows a June suspension driven by national security concerns over potential jailbreaks and unauthorized access by foreign actors.
The Trump administration is allowing Anthropic to release its cybersecurity-oriented Mythos 5 model to over 100 approved U.S. companies and government agencies. This partial reversal of a previous ban includes access for non-American employees at these organizations, signaling a shift in the government's approach to restricting high-capability AI models.
