Changpeng Zhao, known as CZ, posted a security alert about SafePal on X, urging the community to stay alert to phishing attacks. In the same update, he disclosed that SafePal is one of the portfolio companies of YZi Labs, formerly Binance Labs. He also said YZi Labs is currently a minority shareholder in SafePal. The post focused on the security warning and the firm’s investment relationship with SafePal. No other details about the incident were provided in the brief update cited by Odaily.

OpenAI has disbanded its Preparedness Team, the group set up to assess major and catastrophic risks from frontier AI models, according to the Financial Times. The change took place at the end of July, with the company shifting work that had been concentrated in one unit to senior staff embedded in existing teams handling areas such as biosecurity and cybersecurity. OpenAI did not end the work itself; it redistributed it. The Preparedness Team was formally launched in October 2023 and was initially led by MIT professor Aleksander Madry. Its remit was not ordinary ChatGPT content moderation. Instead, it evaluated whether advanced models had reached capability thresholds that could cause serious societal harm. OpenAI had listed core research areas including cyberattacks, chemical, biological, radiological and nuclear risk, large-scale persuasion and manipulation, and AI self-replication and adaptation. The group also ran capability testing and red teaming on unreleased models. The move fits a broader pattern inside OpenAI. The company has repeatedly dissolved or reorganized standalone safety groups, arguing that safety should be integrated directly into research and product development rather than handled as a separate final check.

SafePal said on Aug. 16 that a security flaw in its order-tracking plugin allowed unauthorized access to some customer order records. The company said around 39,798 users were affected, all of whom placed orders between March 2, 2025, and April 11, 2026. The exposed information included names, email addresses, shipping addresses, phone numbers, and purchase details. SafePal said wallet data remained secure throughout the incident. The company added that mnemonic phrases, private keys, wallet passwords, bank account information, payment card numbers, and government-issued identity documents were not involved. According to the announcement, the issue has been fixed and additional security measures have been added. All affected customers have already been notified individually by email. SafePal also launched an official verification page that lets users check whether they were affected by entering their order number and shipping country. The company apologized for the incident and urged users not to share mnemonic phrases, private keys, or passwords with anyone, while staying alert to phishing attempts and impersonation scams. SafePal said follow-up updates will continue to be posted on its official blog.

OpenAI is going through a fresh round of leadership and structural changes just two months after confidentially filing a draft S-1 with the U.S. Securities and Exchange Commission. Between Aug. 11 and Aug. 13, former COO Brad Lightcap said he was leaving after eight years at the company, while Chief Revenue Officer Denise Dresser also departed after only eight months in the role. OpenAI named Dali Rajic, previously at Wiz, Zscaler and AppDynamics, as its new CRO. Separately, longtime technical contributor Scott Gray changed his X bio to describe himself as a former OpenAI GPU specialist, adding to concerns about talent churn. According to the Financial Times, OpenAI has gone through nearly six reorganizations this year, while the company also dismantled its standalone Preparedness team in late July and folded safety work into existing structures. At the same time, the business is shifting fast toward enterprise customers. CNBC reported OpenAI’s annualized revenue has reached $40 billion, with enterprise growth outpacing consumer revenue, while Anthropic said its own annualized revenue topped $47 billion earlier this year.

Anthropic said in a company-wide risk report published on Aug. 14 that it raised its rating for catastrophic harm caused by misalignment in high-risk scenarios from “very low” to “low” under its Responsible Scaling Policy, RSP v3.4. The company said the change was not triggered by a model failing safety tests. Instead, it pointed to recent cybersecurity evaluation disclosures that increased uncertainty, along with a more technical issue: the internal benchmark it uses to detect whether models have crossed the most dangerous capability threshold has become saturated, meaning scores have effectively hit the ceiling and can no longer measure incremental gains in capability. The same report also disclosed an internal system called Model 2 for the first time. Anthropic said the model is slightly more capable than its frontier model Mythos 5 and is already used extensively inside the company. At the same time, it has not yet completed the full set of pre-deployment evaluations that would normally be required before release, and there are currently no plans to launch it externally. The disclosure comes as Anthropic is pushing toward an IPO, putting fresh attention on the gap between frontier-model capability and the tools used to evaluate safety.

OpenAI alerted the FBI after a Goldman Sachs analyst used ChatGPT to describe a plan to kill his ex-girlfriend and uploaded photos of three firearms, according to details cited in the case record. The suspect, 25-year-old South Palm Beach resident Darren Zhou, was later arrested after the evidence was passed to local law enforcement in Palm Beach County. Prosecutors filed three charges in June 2026, carrying a combined maximum sentence of 25 years, and Zhou was fired by Goldman Sachs the same month. On Aug. 12, he pleaded guilty to all three counts under a plea agreement. A judge later sentenced him to eight years of probation, with the first two years under electronic monitoring, and ordered counseling, a mental health evaluation, and restrictions on weapons, alcohol, drugs, and contact with the victim. The court also used a withhold of adjudication disposition, meaning he would avoid a felony record if he complies with probation terms. The case has renewed scrutiny over OpenAI’s reporting standard after an earlier Canadian shooting in which ChatGPT had banned the shooter’s account months before the attack but did not notify police.

Foresight News’ latest weekly Web3 roundup pulls together a dense mix of stories that shaped industry conversation over the past several days. At the center were the death of Quantum Fintech Group founder Harry Yeh in Paraguay, a fresh count showing that more than 300 Web3 projects have gone quiet over roughly 588 days, and renewed attention on Li Lin’s return through Avenir Group-backed UMX. Bitcoin also saw an old fault line reopen after BIP-110 triggered a chain split and cost Luke Dashjr his BIP editor role. The security section was just as busy. A DeFi whale lost roughly $25 million to $26 million in minutes, Trezor disclosed a ShipMonk-related data breach affecting close to 14,000 customers, Harmony was hit by an exploit that expanded ONE supply, and ZachXBT detailed a social-engineering scam tied to at least $5 million in losses. Foresight also reviewed Bybit’s effort to recover funds from the 2025 theft attributed to Lazarus Group, with recovered and frozen assets totaling about $78.9 million so far. On the project side, Vitalik Buterin’s updated Ethereum Strawmap elevated privacy and post-quantum security, Justin Drake said Ethereum would drop Poseidon at L1, Bitwise continued launching products while shutting eight ETFs and cutting staff, and Trump Media & Technology Group showed how far it has moved beyond a pure social-media business.

Trezor disclosed on Aug. 13 that its logistics partner ShipMonk suffered unauthorized access to a system holding customer information, affecting 13,689 people. The exposed data fell into two groups: 11,742 customers had their full names, email addresses, phone numbers and full shipping addresses exposed, while 1,947 had their names, city and email addresses exposed. Trezor said its own systems, products and services were not compromised, and that hardware wallets, private keys and wallet backups were not part of the incident. The disclosure came two days after a user posted that the outer label on a domestically shipped Trezor Safe 3 package explicitly read "Trezor Safe 3 Bitcoin Only" instead of a generic product description. There is no confirmed direct link between the shipping-label complaint and the ShipMonk breach, but together they highlight the same problem: a hardware wallet may protect private keys, while purchase, fulfillment and delivery processes can still connect the device to a real-world identity. The episode has also fed a broader debate over hardware-wallet risk. Recent incidents involving Trezor, Coldcard and Ledger did not stem from the same failure point. Some relate to device or firmware security, while others involve third-party commerce, logistics or packaging exposure. For affected users, the most immediate concern is targeted phishing and social-engineering attempts that use real names, addresses, order details and device information to appear credible.