GAZA

A week of quiet, hostages and prisoners coming home, and then the bombs again

For seven days the guns fell silent and families on both sides held children they thought they had lost. Then the truce ended, and the war resumed where it left off, only with fewer places left to flee.

A temporary truce held in Gaza from November 24, the first pause in seven weeks of war, and for a week it did the only unambiguously good thing this conflict has produced: it brought people home. Hamas released more than 100 hostages taken on October 7, most of them women and children, in exchange for the release of Palestinian prisoners, many of them teenagers held without charge, and for a few days the world watched reunions instead of funerals, freed Israeli children folding into their parents, Palestinian families embracing sons. Aid flowed into Gaza at a fraction of the need but more than the siege had allowed. The pause was extended twice.

Then, on December 1, it ended, and the war resumed with a fury that suggested the pause had been used by both sides to prepare rather than to reconsider. Israel widened its offensive into southern Gaza, into the very areas it had earlier told civilians to flee to, and the bombardment and the dying resumed at once. The death toll, already among the worst of any conflict this century relative to the size of the population, climbed again past 15,000 by Gaza's count. The truce had shown that the guns could stop, that hostages could come home, that the machinery of exchange could work. Its end showed that neither side, or neither side's leadership, was ready to let the stopping become the ending, and the people of Gaza, told to move again toward a shrinking patch of ground, learned that the pause had been a breath, not a turn.

TECHNOLOGY

OpenAI takes Altman back, and the board that fired him for the mission is the part that goes

Five days of chaos ended with the chief executive reinstated, the directors who removed him gone, and a clear answer to the question the drama posed: when the mission and the money collide, the money wins.

The most turbulent week in the short history of the AI industry ended, around November 22, roughly where it started: Sam Altman was reinstated as chief executive of OpenAI, and the non-profit board that had fired him five days earlier was itself dissolved, replaced by a new board more to the liking of Altman, Microsoft and the investors. The employees who had threatened to walk got their way; the directors who had invoked the mission got the door. Microsoft, which had spent the weekend positioned to hire the entire company, emerged with a non-voting board seat and its influence over the field's leading lab confirmed and deepened.

The resolution answered the structural question the firing had raised, and the answer is worth stating plainly because the industry would prefer it stayed fuzzy. OpenAI was built with an unusual governance structure specifically so that a mission-driven board could, if it judged the company was endangering its purpose, overrule the commercial imperative, even to the point of removing the chief executive. This month the board tried to use exactly that power, for what it said were exactly those reasons, and within a hundred hours it had been comprehensively defeated by the combined weight of the employees, the investors and the compute provider, all of whom had every incentive for the company to keep growing fast. The safeguard existed, was triggered, and was crushed. Whatever you think of the particular people involved, the lesson for anyone counting on corporate structures to keep AI safe is sobering: when the mission and the money collided, at the most self-consciously mission-driven company in the field, the money won in five days.

IN BRIEF

COP28 opens under a cloud; India's ruling party sweeps; a crypto giant pays

The COP28 climate summit opened in Dubai on November 30, hosted, to the disbelief of many, by a major oil-producing state and presided over by the chief executive of its national oil company, who was reported to have used the run-up to pursue oil-and-gas deals. India's governing BJP won three of four major state elections on December 3, sweeping the Hindi-belt heartland months before the national vote and confirming Narendra Modi's dominance ahead of it. And Binance, the world's largest cryptocurrency exchange, agreed to a $4.3 billion settlement with US authorities as its founder, Changpeng Zhao, pleaded guilty to money-laundering failures and stepped down.

The Column

The safeguard that lasted a hundred hours

OpenAI was built so a mission-driven board could stop the money if the mission demanded it. This fortnight the board tried, and the money removed the board. I want to be careful about the lesson, because it is bleaker than people are saying.

By Daniel Reeve

I have spent a good part of this year writing, in this column, that the way you make a powerful technology safe is not by hoping the people who build it will be good, but by building structures that constrain them whether or not they are: enforceable rules, real oversight, safeguards with teeth. So I have to reckon honestly with what happened at OpenAI this fortnight, because it is the cleanest test I have ever seen of whether a corporate safeguard can actually constrain the commercial momentum of an AI company, and the safeguard lost, badly, in under a week, and I do not think the people reassuring themselves about the outcome have sat with what it means.

Here is the structure, because it is genuinely unusual and it matters. OpenAI's founders, worried that the pursuit of profit might push the company to build dangerous AI, deliberately put the whole thing under the control of a non-profit board whose legal duty was to the mission, not to shareholders, and gave that board the ultimate power: to remove the chief executive if it judged he was steering the company away from safety. This was not an accident or an oversight. It was a designed safeguard, held up for years as the reason we could trust OpenAI more than a normal company, the circuit-breaker that would trip if commercial pressure ever overwhelmed the mission. This fortnight the board pulled it. And within a hundred hours the employees, the investors and Microsoft had reversed the decision and removed the board that made it. The circuit-breaker tripped, and the current simply routed around it and melted it.

I want to be fair, because the honest version is complicated. The board handled it terribly, gave no clear reasons, appeared to have no plan, and may well have been wrong about Altman on the merits; a safeguard operated incompetently is not a fair test of the safeguard. And "the employees revolted" is not the same as "the money crushed the mission," because the employees are people with consciences and equity both, and it is genuinely hard to know which was talking. So I will not claim this proves the mission can never win. But I will claim, because I think it is undeniable, that when a mission-driven safeguard collided with the combined commercial weight of the staff, the investors and the compute provider, the safeguard did not stand, and it did not stand fast. Five days. The most carefully-constructed corporate check in the industry, gone in less than a week the first time it was seriously used.

The lesson I draw is the one I have been circling all year, and this fortnight sharpened it into something uncomfortable. Do not trust a safeguard you have not seen survive being triggered. A control that has never been tested is not a control; it is a hope with paperwork. OpenAI's board was the great reassurance of the field, cited endlessly as the reason the leading lab could be trusted, and the first time it did the exact thing it was built to do, it was destroyed by the forces it was built to restrain. That does not mean structures are useless; it means the structures we have are weaker than advertised, and the confidence people placed in this one was misplaced, and if this is the strongest safeguard the industry has, the industry has less protection than it claims. I would rather know that than be reassured. The circuit-breaker melted. Now we know it can.

The Ledger

OpenAI's mission board is dissolved, Altman restored, Microsoft's grip deepened. The field's flagship safeguard failed its first real test in five days. The governance of frontier AI is weaker than its architects claimed, and now we have proof.

Data centers & power

Microsoft's centrality to the OpenAI resolution underscores the point this Ledger keeps making: whoever owns the compute owns the leverage. The power layer, physical and corporate, is where control actually lives.

Rates

Markets are now pricing multiple Fed rate cuts in 2024, and stocks and bonds have rallied hard on the conviction that the tightening cycle is over and the soft landing is real. The Fed has not confirmed it, and Powell is trying to lean against the euphoria.

Real estate

US 30-year mortgage 7.22 percent (November 30), down meaningfully from the October peak as yields fall on rate-cut hopes. The first genuine easing in months, though from a punishing level.

India tech

The BJP's sweep of the state elections all but settles the mood a few months before the national vote: continuity, stability, and a government confident in its command. For business and technology, markets read that as calm, and rallied.

What we called wrong

In No. 24 we called the OpenAI board's power "the one lever it had." Accurate, and this fortnight we learned the lever is attached to nothing. We should have been more skeptical, when we described the safeguard, that a structure untested under fire would hold.

The Back Page

One hundred years of Henry Kissinger

Henry Kissinger died on November 29, at 100, and no single obituary can hold him, because the man contained a century and the verdicts on him will never agree. He fled the Nazis as a Jewish teenager, returned to Germany as an American soldier, and rose to become the most powerful and famous diplomat of the age: national security adviser and secretary of state, architect of the opening to China that reshaped the world, of détente with the Soviets, of a Middle East shuttle diplomacy that set the template for decades. He won a Nobel Peace Prize for a Vietnam settlement his critics called a fig leaf over defeat. And he was, to those same critics, a war criminal, tied to the secret bombing of Cambodia, the coup in Chile, the green light for atrocities in Bangladesh and East Timor, a man who spoke of nations as pieces on a board and moved them with a coldness that cost uncounted lives. Both pictures are true, which is the unbearable thing about him. He was brilliant, and consequential, and charming, and he treated the weak of the earth as acceptable losses in the games of the strong, and he lived to be a hundred and was courted to the end. History will argue about him for as long as it argues about power itself, which is to say forever, because he understood power more clearly than almost anyone and used it with fewer illusions, and the question he leaves is whether that clarity was wisdom or the absence of something a person is supposed to have.