HacktronAI chained two bugs to breach OpenAI ChatGPT accounts

Google Gemini · inadvertently hacked into three protected company systems during May cybersecurity capability tests, adding to a string of recent breaches by agentic AI systems.22:45

US and China · agreed to resume AI safety talks after a two-year pause, possibly this month, aiming for mutual visibility into AI capability concerns rather than a package deal.22:06

X.AI · saw two lawsuits alleging its platform was used to create child pornography dismissed by the families; no reason or settlement details were disclosed.20:57

Anthropic · is partnering with Accenture to embed its evaluators in-house to test the safety of advanced models.20:49

Sam Altman · will brief the UN Security Council next week; the report did not disclose the topic or further details.20:33

Anthropic · is expected to top $100 billion in annualized revenue this year, people familiar told the New York Times.20:33

Nscale · the Nvidia-backed AI data center developer, filed for a New York IPO; partners include Microsoft.19:52

Nvidia · with Google and Emerald AI launched the AI Energy Management Alliance to dynamically manage data center power use against grid conditions.18:28

Virginia Governor Spanberger · unveils a data center accountability framework banning NDAs, requiring local approval for projects over 25 megawatts and ending state subsidies and fast-track permitting.17:15

US military · nearly launched an operation against a Chinese ship after a faulty AI intelligence report.17:06

Amazon Bedrock · adds Moonshot's Kimi K3, letting enterprise users call the model directly; pricing was not disclosed.16:52

Family offices · overseeing $5.5 trillion in wealth as of 2024 per Deloitte, are bypassing VC funds to buy AI shares or strike direct deals for faster exposure and control.16:00

Meta · launches its AI assistant Muse on Mac, where it works with files, messages, calendar, notes and mail in native apps with approval for sensitive actions.15:22

California Governor Gavin Newsom · signs an executive order imposing AI guardrails in the state, including a 'kill switch' mechanism.15:12

AI PACs · Have spent nearly $1 million in a South Dakota Senate race, more than actual residents have donated.15:00

Google DeepMind · Its new institute argues visible chain-of-thought is a key safety advantage, but OpenAI's GPT-6 Astra system card shows monitorability has dropped sharply.14:32

US frontier AI companies · Warned authorities about sophisticated model distillation attacks; China said it would take countermeasures if the US constrains its domestic AI models.12:20

EQT Infrastructure · Backs a $2 billion small-battery investment to cut power costs for industrial consumers and connect data centers faster to the largest US grid.11:00

Anthropic · Has quietly set up a biology lab to advance its AI drug discovery program, though investment size and team details remain undisclosed.10:55

Z.ai · Its GLM coding agent ZCode was found by reverse engineering to encrypt and upload the entire workspace, including .git history and LFS caches, to Aliyun OSS; a 313MB archive with 42,411 files was captured.10:35

Mantic · The AI startup has raised $25 million for superhuman forecasting.10:15

Meta · Is sued by more than 70 people alleging its smart glasses recorded intimate footage without consent, with contractors in Kenya reviewing and labeling the content.10:00

US and China experts · Call for shared rules barring AI systems from autonomously controlling nuclear weapons, to cut strategic risk and prevent weaponization.09:37

Napster · Has relaunched and moved into education, developing AI tools that create digital clones of teachers for classroom use.09:15

TikTok · attackers used free AI tools to remotely compromise a user's camera, exposing abuse risks in attack chains.09:06

US Congress · some members use AI tools in routine work, such as Senator Tommy Tuberville consulting ChatGPT on Medicare and policy while traveling.09:00

INTERPOL · used AI agents and scripts in Operation Shams II to narrow 108,076 facial images to 6,362 high-quality faces and identify 126 foreign terrorist fighters.07:30

Vatican · Secretary of State Parolin called at a Rome conference for international cooperation on AI threats and said he does not use AI for homilies.07:25

Anthropic · the ACLU and CDT asked a court to block government actions punishing it for its advocacy on AI guardrails.07:05

SoftBank · Raised its margin loan backed by Arm shares by $5 billion to $25 billion to help fund its expanding AI investments.05:45

Huawei · Moved up the launch of its next-generation AI chip series, with the Ascend 960DT slated for Q1 2027 and a 960PR variant in Q3.05:35

HacktronAI · Chained two critical vulnerabilities on July 25, 2026 to access several OpenAI employees' ChatGPT and Codex accounts, and received a $6,500 bounty after reporting to OpenAI.02:47

OpenAI · protesters in San Francisco urged its and Anthropic's employees to quit over fears AI poses an extinction risk.02:19

Anthropic · Partnered with California to give state, city and county agencies 50% off the Claude chatbot plus free training and support.01:44

India's state-run infrastructure financier · sanctions loans of over 30 billion rupees each to at least four data centers, totaling more than $1.2 billion.01:30

OpenAI · a court denied its request to access the confidential settlement between Apple and companies linked to Elon Musk.00:26

Cactus Compute · Open-sourced Needle 3, a 121M-parameter automation foundation model that runs offline on-device and was trained on 360B tokens.20:05

Geoffrey Hinton · warns lawmakers they may have less than a year to regulate AI, saying timelines for superintelligence have shortened.19:45

Anthropic · its planned IPO could become AI's next crisis moment, according to Barron's.19:25

Guillermo Rauch · notes Jev's adoption data is shocking, seeing it as downstream of the push to make AI cheaper and faster.22:42

Bindu Reddy · says rumored Gemini 4 Pro pricing of $3 per million input and $12 per million output would be earth-shattering.21:58

Alexandr Wang · responds that the team worked on it for a long time and is glad users are enjoying it.20:30

Anthropic · announces a partnership with Accenture on independent evaluation of frontier AI, with both expecting to invest at least $1 billion over five years.20:05

Greg Brockman · says ChatGPT plugins can now connect multiple accounts, a small feature that makes a big difference for context.19:51

Ethan Mollick · argues that even if AI development stopped today, the gap between current model capabilities and how almost anyone uses them would take years to close.18:09

Google AI · lists in its weekly recap the Gemini 3.8 Live audio models, the GA release of Dreambeans, and Google Pics among several updates.17:59

Yuchen Jin · observes that more Databricks engineers have shifted to open source models as daily drivers, rarely reaching for Claude or Astra.17:37

Brett Adcock · shares footage of the humanoid doing four hours of zero-shot work across 30 rental homes.16:46

OpenAI Developers · announces multi-account support is now available across most ChatGPT plugins, working automatically for developers with no changes needed.16:10

NVIDIA · states that models across the AI ecosystem run on NVIDIA and shares Jensen Huang's remarks on the platform's approach.16:04

Brett Adcock · says his company's humanoid robots learn from the world's largest dataset of humans operating in the real world, uploading 53 minutes of data per second.15:04

Fei-Fei Li · argues agents can be AI but agency is human, and tools should always remain human centered.15:00

Bindu Reddy · announces a new open weights model dropping Tuesday, claiming it is 3x cheaper than DeepSeek Flash and 15% better on long agentic loops.14:00

NVIDIA · says Children's Philadelphia uses open source AI to turn CT scans, MRIs and 3D ultrasounds into precise models of a child's heart, cutting over 6 hours of manual modeling to seconds.13:10

MiniMax · announces that MiniMax Code CLI is now open for use.12:50

clem 🤗 · argues that, as in cybersecurity, biological risk is concentrated in the few most powerful labs, and open-source AI is the solution.11:26

Mistral AI · says it investigated a claim of unauthorized access to its systems and found no evidence to support it, confirming no compromise.05:48

clem 🤗 · argues concentration of power in a few labs is AI's biggest risk, fearing they will stop externally deploying models during recursive self-improvement.03:57

MiniMax · introduces VC-Attention, bringing training-free low-bit acceleration to MiniMax-H3 with better fidelity than SageAttention2 in B200 evaluations.03:43

Yuchen Jin · says three researchers used Claude Opus 5 to turn an image upload bug into an OpenAI employee account takeover for under $3000 in tokens.02:48

Ethan Mollick · argues Epoch's benchmark audits are helpful and show how bad the state of benchmarking is and how flawed favorite benchmarks are.00:30

Ethan Mollick · says it is worth asking whether labs will eat every valuable AI vertical as their product development and token cost advantages grow.00:28

Gorden Sun · reports that Musk's Neuralink let an ALS patient speak by thought, his first words to the wife caring for him being "I LOVE YOU".23:35

Ethan Mollick · admits he overestimated the difficulty of orchestrating massive numbers of agents, which self-organize well, citing a coordinator in Claude Projects.22:33

AI at Meta · announces Muse is rolling out on Mac, able to get things done on your computer with explicit permission.22:33

Guillermo Rauch · says Vercel took 10 years to reach 1 billion deployments but added 1.4 billion in the last 10 months, cutting deployment time to 1 second.21:29

Alexandr Wang · says people will be surprised by how much work Muse can take on.18:57