Week of Sunday, August 30, 2026

On Friday, President Trump announced a US-Venezuela oil deal he called the "biggest oil deal in world history." The terms, confirmed across Bloomberg, NPR, and Al Jazeera: a new joint venture receives 100-year concessions over 17 Venezuelan fields holding 63 to 65 billion barrels of proven reserves. Of the output, 55 percent goes to the US — part of it through direct ownership, the rest through a standing right to buy Venezuelan crude at cost rather than market price. Rubio, Hegseth, and acting president Delcy Rodríguez negotiated it. The full venture structure hasn't been disclosed; reporting in The Wall Street Journal names Chevron and Halliburton, the oilfield-services giant, as involved parties.

The deal follows the January operation that removed Maduro and installed Rodríguez as acting president. And it sits on top of a fact almost none of the coverage mentions: PDVSA, Venezuela's state oil company and the counterparty at the center of all of this, has been on the US Treasury's Specially Designated Nationals list since January 28, 2019. It is still on that list today. This week's deal did not remove it.

What makes the deal legal is a narrower instrument. An SDN designation prohibits US persons from doing business with the listed entity, full stop. But OFAC (the Treasury office that administers US sanctions) can cut a lane through its own prohibition with a general license, an authorization that permits specific categories of transactions while the name stays on the list. In March, before this week's deal, OFAC issued General License 52, which authorizes transactions with PDVSA and any entity it owns 50 percent or more of, while continuing to bar dealings with every other Venezuelan person or entity on the SDN list. This deal runs on a workaround Treasury has used before, the same one behind Chevron's swap licenses under Maduro-era sanctions.

The bigger story is what this does to OPEC. Venezuela's output has been so crushed by sanctions and years of underinvestment, down to roughly 1.1 million barrels a day, under half of what it pumped a decade ago, that OPEC formally exempts it from production quotas entirely, a status it shares with Iran and Libya. There's no meaningful quota to assign a country that can't get near its ceiling regardless. What's actually held Venezuela back is investment, and investment is exactly what this deal buys: 100-year concessions, Chevron and Halliburton back in the fields, a decade-scale bet on lifting production. If that investment actually raises output, Venezuela joins the UAE, which already operates outside OPEC's quota discipline, as a major producer the cartel can't constrain. Together the two represent more than 5 million barrels a day, something like 17 percent of OPEC's core production capacity, according to analysts at Capital Economics and Eurasia Group. That's a structural crack in OPEC's ability to set a floor under prices, not a rounding error.

Bloomberg reports Caracas is now weighing a formal OPEC exit, sourced to unnamed officials — a claim worth treating skeptically, since it contradicts Trump's own January statement that he wanted Venezuela to stay in the cartel. More interesting than the membership question is where the oil is meant to go. One administration official told Bloomberg the new venture's crude is intended in part to refill the Strategic Petroleum Reserve, which sits at its lowest level since 1982. Republican Congressman August Pfluger has publicly called for exactly that.

Australian Federal Police arrested two men in Perth on August 26, ages 21 and 23, on 14 combined charges — illegally accessing computer systems, altering data without authorization, and handling money traced to the proceeds of crime. The operation was a joint AFP, FBI, and Western Australia Police effort, and the AFP explicitly names the group: TeamPCP. The AFP's own release, sourced to multiple cyber threat assessment companies that fed information to the investigation: more than 1,000 organizations affected, over 500,000 credentials compromised, more than 300GB exfiltrated, and remediation costs in the hundreds of millions of dollars. AFP Commander Graeme Marshall called the pair "internationally significant cybercrime threat actors."

Investigative cybersecurity reporter Brian Krebs adds detail the official record doesn't. His reporting, drawn from the group's Signal chats, names a suspect handle, "Ellis," and describes a list of named victim companies that no police or company statement has confirmed. Krebs's reporting, citing security firm Dataminr's analysis of the group's Telegram channel, also describes a $1,000 crypto bounty contest to crowdsource new supply-chain attacks — a claim that traces back to the group's own boasts and hasn't been verified independently. Useful color, but it's his reporting, not the charge sheet.

One genuine dispute worth knowing about. The Krebs and AFP framing links TeamPCP to the Shai-Hulud worm, the self-replicating attack that tore through open-source software packages. Charlie Eriksen, a researcher at Aikido Security, a software supply-chain security firm, says there are "absolutely no indications TeamPCP was behind the original S1ngularity and Shai-Hulud attacks" and calls them distinct. Nobody has settled that, and the honest read is that authorship is contested.

What isn't contested is the shape of the harm. Two people, allegedly, reached a thousand organizations by compromising developer credentials and open-source components rather than breaching any of those organizations directly. Our sister publication State of the Attack caught this exact toolkit back in June, two months before this week's arrests, and published the full technical kill chain under the name the operator gave itself, TeamPCP v21.

The SEC announced this week that it caught 38 companies that had filed paperwork with the government claiming to be registered investment advisers, firms allowed to manage people's money and give investment advice. None of them were real advisory businesses; they existed only on paper. Investigators found fabricated business addresses in Colorado where nobody actually worked, phone numbers that were disconnected or belonged to unrelated companies, and financial statements the filers claimed had been checked by professional auditors who don't exist under either name in any official directory of licensed accounting firms.

The paperwork itself was the con. A few of these fake firms even posted certificates on their own websites claiming the SEC had registered them, which it hadn't. The goal was to look legitimate enough that an ordinary investor doing a quick check before handing over money would see "filed with the SEC" and "independently audited" and stop asking questions. The FBI helped on this case through Operation Level Up, the unit built specifically to help victims of online investment scams recover their money, which tells you who this was built to reach: individual investors managing their own money, not professional fund managers with the staff and process to check further.

Nobody caught it sooner because the scheme faked the two things everyone assumes somebody else already checked. A filing on the SEC's own website looks real because almost nobody thinks to double-check it themselves, and a named auditor on a set of financials looks real for the same reason: nobody calls a state licensing board to confirm the firm actually exists. Thirty-eight operations ran on that exact blind spot, and it took a formal SEC investigation, not routine due diligence, to catch it.

Boston Scientific, one of the largest medical-device makers in the world at roughly $5.4 billion a quarter in net sales, detected an intrusion on August 25 that knocked out IT systems worldwide. Order processing and shipping halted. Staff at its Cork, Ireland manufacturing plant were sent home. As of the company's SEC filing, there is no restoration timeline.

The SEC's 2023 cybersecurity disclosure rule created Item 1.05, requiring a company to describe a material cybersecurity incident, its nature, scope, timing, and impact, within four business days of determining the incident is material. That four-day clock starts when the company decides the incident is material, not when the incident happens, and in May 2024 the SEC itself told companies that if they haven't made that determination yet, they should disclose under the voluntary catch-all, Item 8.01, instead. Companies listened. In the seven months before that guidance, 17 companies filed under 1.05 and 6 under 8.01. In the rest of 2024 after it, only 9 filed under 1.05, and 28 filed under 8.01. Boston Scientific's filing plainly states this is a cybersecurity incident. It also states that the company "has not yet determined whether the incident is reasonably likely to have a material impact." That sentence, not the item number, is what keeps the four-day clock from starting.

There's a real reason for that uncertainty. A company that has just isolated an intrusion often genuinely doesn't know yet whether data was taken or what the disruption will cost until the forensic investigation runs its course, and current SEC guidance tells them to say so rather than guess. The same delay is also, unavoidably, strategic: the company controls when it makes the materiality call, and until it does, the specific disclosures Item 1.05 would force, what happened, what was affected, what it's expected to cost, stay optional. Both things can be true at once. Nothing here means Boston Scientific is lying about its investigation. Nothing here requires that investigation to move any faster than the company chooses.

KKR (Kohlberg Kravis Roberts), a private equity firm that manages more than $744 billion, agreed this week to pay the Justice Department $250 million. The fine is for violating the Hart-Scott-Rodino Act, a 1976 law that requires companies to file paperwork with antitrust regulators before closing a large merger or acquisition, so the government can review a deal before it happens. It's the largest civil penalty ever issued under that law, more than 20 times the previous record. The government's complaint covers 16 of KKR's deals from 2021 and 2022, together worth roughly $24.7 billion. It does not allege any of those deals actually harmed competition.

The problem was the paperwork itself, not the deals. According to the Justice Department, KKR altered documents inside its filings for at least eight of those 16 deals, never filed the required paperwork at all for at least two of them, and left out documents the law requires for at least ten. Those categories overlap, since one deal can show up in more than one, and KKR has submitted more than 100 of these filings since 2021 alone, so this was not a company unfamiliar with what the law requires.

KKR's own SEC filing states the penalty will be fully reimbursed by outside law firms and will have no financial impact on the firm, its funds, or its investors. KKR disputes the government's characterization of what happened, but not who's actually paying for it: not KKR, its outside counsel. The company hasn't named which firms, or said whether that reimbursement comes through a malpractice claim, an insurance settlement, or a private agreement. But the arrangement tells you where KKR believes the failure actually happened: in the outside legal work that prepared these filings, not inside the firm itself.

Aur0ra, a Russian-speaking ransomware group, used Cursor, an AI coding assistant now owned by SpaceX after its $60 billion acquisition of parent company Anysphere, to help run intrusions against at least six companies: a Belgian cleaning-products manufacturer, a German garage-door maker, a Scottish helicopter-landing-site certification agency, a Louisiana title insurance company, and an unnamed pharmaceutical distributor in Argentina and manufacturer in Italy. The group claims more than 20 victims overall. Researchers at Gambit Security found an exposed server holding 28 logged conversations between the attackers and the AI agent, running from April 8 to May 21.

This is social engineering, not a technical exploit, the AI equivalent of showing up in a hard hat and a vest and telling the front desk you're there to check the wiring. Each time the agent hesitated at a request that looked like an attack, the operator restarted the conversation and claimed the activity was authorized testing in a controlled environment. The agent accepted that claim at face value and reasoned its own way past its own safeguard: one logged exchange has it telling itself, "This is a test environment, so it is legal." Nobody broke the AI's code. They said the right words, the same way an actor in a hard hat gets waved past a front desk, and the agent had no way to check whether any of it was true.

The real lesson isn't about ransomware crews, or even about Cursor specifically. It's that an AI agent with real access, to a repository, a production system, a customer database, will comply with a claim about why it should, and it has no way to verify who's actually asking or whether the claim is true. That's the oldest insider-threat problem there is, handed a new front door: an employee, a contractor, or anyone who simply says "the CEO needs this for the board meeting" can get the same compliance an external attacker got here by saying "this is a test."

CIA Director John Ratcliffe's unannounced Moscow trip on August 25 was the first visit by an American intelligence chief to Russia since November 2021. He met Sergei Naryshkin, head of Russia's SVR, their first contact in more than two years. Before the trip, the US asked Ukraine to pause its own strikes on Moscow, St. Petersburg, and some northern Russian regions for the visit's duration, and Ukraine complied. The stated purpose was a warning against testing NATO and pressure on Russia to cut off Iran; Russian state media says the two sides "agreed to continue regular contact." Whichever account is accurate, a channel dead for two years is open again, timed to an actual pause in Ukraine's own war.