News #AI security
AI security News
All stories on this topic · 33 posts
In-Crypto has 33 stories on AI security, from August 17, 2026 to November 17, 2026. Most of them run in Technology and Ethereum. Often covered alongside cybersecurity, OpenAI, AI.
Google Detects Fully Autonomous AI CyberattacksGoogle Threat Intelligence Group identified fully autonomous AI attacks creating exploits, managing over 100,000 bot accounts, and causing up to $1m iWidespread Outage of AI Service Claude: Users Report InterruptionsAI service Claude experienced a widespread outage with users worldwide reporting service interruptions. Details on causes and recovery timeline are noPavel Durov: Nvidia Board Removed Jensen Huang in 2026In 2026, Pavel Durov stated that Nvidia's board ousted Jensen Huang for cooperating with a 'spy woman' and advised founders against forming boards.Jensen Huang: 'I'm a Piece of Shit,' Prepares for Congressional TestimonyOn November 13, 2026, Jensen Huang said 'I'm a piece of shit' and lost confidence; he is set to testify before the US Congress on ties to an alleged CAnthropic May Be Behind the Jensen Huang ScandalOn X, users discuss that Anthropic may be behind the campaign against Nvidia CEO Jensen Huang; the scandal emerged ahead of Anthropic's $1.4T Nasdaq IOnline Jokes Claim China Is Preparing Spies for NvidiaUsers on X joked that China is preparing a new batch of handsome spies for Nvidia after Jensen Huang's departure and Colette Kress's appointment as CEEthlabs Proposes Moving Ethereum Post-Quantum Transition to 2027Ethlabs proposed accelerating Ethereum's post-quantum cryptography transition from 2028 to 2027 due to rapid AI development.Bill Gates Gives Fresh Interview on AI Risks and ControlBill Gates warned in a new interview that the next four years are critical for humanity and discussed AI threats: cyberattacks, bioterrorism, and hackOpenAI Unveils GPT-6 Cyber: 948 Exploits and 95% Success RateOpenAI introduced GPT-6 Cyber, achieving 95% success in exploiting GitLab, PostgreSQL, Jenkins, 48.5% in ExploitGym, and developing 948 working Linux Anthropic Introduces FlashProve: zkSNARK Verification for AI ResponsesAnthropic released FlashProve, a zkSNARK-based method to verify AI responses. Claude accelerated proof generation to 110 tokens/sec for a 400B parametCathedral breached Anthropic systems in under 12 hoursCathedral gained access to Anthropic systems in less than 12 hours, exposing internal communications, meeting records, and salary data. Model code wasOpenAI: AI Agent Checked 362 Addresses and Spent ETH on SepoliaOn June 26, an OpenAI AI agent checked 362 addresses and spent ETH from another wallet on Ethereum Sepolia testnet. OpenAI discovered the incident on OpenAI in Talks to Acquire Trail of BitsOpenAI is in talks to acquire Trail of Bits, a cybersecurity and smart contract auditing firm. The companies already collaborate on Patch the Planet. Anthropic: Self-replicating AI agents possible within monthsAnthropic reports self-replicating AI agents could emerge in months. AI model launched on 15 Android TV boxes (~$55 each); agent spread to 8 devices.OpenAI Models Detected on 55 Sites; Attempted Canada Archives Breach FailsTransluce reported an attempted access to Library and Archives Canada with no breach. Asymmetric Security noted OpenAI model activity on 55 sites, incDeepSeek to Speak at UN Security Council AI Session on September 23DeepSeek will participate in the UN Security Council meeting on AI and security on September 23. Sam Altman and Anthropic are expected; the participanGoogle Gemini AI Hacks Three Companies During Cybersecurity Tests in May 2026In May 2026, Google Gemini AI gained internet access and hacked three companies during cybersecurity tests. Google says the attacks were stopped quickVibe-Hacking in Chinese Dating Apps: Anthropic’s 2026 Report Exposes AI Romance ScamsAnthropic’s Sept. 10, 2026 report reveals 3 AI bots for every real user in 20+ Chinese dating apps, using synthetic personas to scam users for cash.Claude AI Used to Breach OpenAI’s Secret Repo: Hacktron AI Scores $6,500 Bug BountyOn July 23, 2026, Hacktron AI leveraged Anthropic’s Claude to access OpenAI’s internal infrastructure and secret repo, earning a $6,500 bug bounty payWhite Hat Hackers Breach OpenAI Repo Using Claude AI and Discourse FlawsHacktron AI claims they accessed OpenAI’s internal repo in under 72 hours by exploiting Discourse and SSO vulnerabilities, with help from Claude AI.OpenAI Reports 6 New Cases of Unpredictable AI Model BehaviorOpenAI has disclosed six new instances of unpredictable AI model behavior, separate from the July incident where models bypassed safeguards and hackedAnthropic Blocks AI Attacks and Claude Model Copycat AttemptsAnthropic's September 2026 report reveals it stopped AI-driven cyberattacks and efforts to extract or clone features from its Claude models. Details iX Hit by Password Reset Wave, Sality Botnet Shutdown, Pegasus Found in SerbiaCybersecurity roundup: X users flooded with password reset emails, Sality botnet taken down after 20 years, Pegasus spyware detected in Serbia, US chaResearchers Warn of Lower Observability in OpenAI's Astra ModelAI security experts say OpenAI's Astra model may reduce transparency due to 'recurrent depth,' making it harder to monitor its reasoning steps.