In The News
The rulebook with an off switch. Homeland Security’s AI rulebook — Directive 139-08, Revision 01, which took effect in September — lets the department’s chief AI officer waive any requirement in the policy, including its own prohibited uses. The prohibition on the page bars AI for “systemic or large-scale monitoring, surveillance, or tracking of individuals, or predictions of emotional state or future behavior” — now subject to that waiver.
Tom Bowman, policy counsel at the Center for Democracy & Technology, told FedScoop (Oct 21): “A rule that comes with its own off switch is not a rule.” Bowman says the chief AI officer can switch the prohibition off with no independent review from the privacy or civil rights offices, no notice to Congress or the public, and no expiration date. Rachel Levinson-Waldman of the Brennan Center’s Liberty and National Security Program calls the waiver “a loophole big enough to drive a truck through.” The directive does say waivers must be “consistent with law, regulation, and government-wide policy” — Levinson-Waldman notes any chief AI officer can assemble a justification explaining away the tension.
The surrounding facts, from FedScoop’s reporting: in September, ICE awarded a $125 million contract to Thomson Reuters Special Services for services enabling monitoring of one million individuals and identifying potentially criminal behavior before it happens; the department’s chief privacy officer no longer must ensure AI “does not erode” privacy protections — now only that it “complies” with guardrails; and language about equity and underserved communities was removed from the directive. Plainly stated: DHS has not said when or whether the waiver would be used.
Italy’s AI law gets teeth — today. Italy’s Legislative Decree No. 179/2026 — promulgated by the President on October 7 and published in the Official Gazette no. 234 on October 8 — takes effect today, October 23, completing Italy’s domestic transposition of the EU AI Act. Overseeing it: AgID (Agency for Digital Italy) as the notifying authority, and ACN (National Cybersecurity Agency) for market surveillance and as the single point of contact with Brussels; the Bank of Italy, CONSOB, and IVASS cover finance and insurance.
The fines, per Italian legal press reporting on the decree text: up to €35 million or 7% of worldwide annual turnover for prohibited AI practices; €15 million or 3% for provider and deployer obligations; €7.5 million or 1% for misleading information to authorities — plus an Italian-only band of €1 million or 0.5% for skipping the fundamental rights impact assessment or an individual’s right to an explanation of a decision. Italian legal press calls that last band the one likely to produce the earliest cases.
The workplace angle: the EU AI Act classes AI used for recruitment, promotion, and dismissal as high-risk; Italy didn’t wait for the EU-wide duties to bite in December 2027 — workers get a concrete remedy now: a void dismissal. No algorithm alone fires a worker. And a criminal layer runs alongside: companion decree 160/2026, in force since September 30, makes failing human oversight on high-risk AI systems a criminal offense — up to five years’ imprisonment, eight where life or safety is endangered, with corporate liability under Decree 231/2001 extended to AI offenses. The contrast: Europe is putting penalties behind its rules while UBS said this week major new US regulation is unlikely — the US states move while Washington doesn’t (California’s No Robo Bosses Act, SB 947, from July 2027; Colorado’s automated-decision rules, January 2027).
Ukraine: the robot guns that “eliminate the human factor.” Ukraine has begun using AI-enabled robot guns to shoot down Russia’s new jet-powered drones, the Ukrainian air force told Agence France-Presse (Oct 21). “Robotic turrets are now being installed on bridges,” air force spokesman Colonel Yuri Ignat said on national television October 5. There have already been shoot-downs, Ignat said — including of the bigger, faster, more powerful Geran-5. The turrets, seen by AFP journalists in Kyiv, use “optical vision and artificial intelligence to eliminate the human factor and deliver strikes.”
Sergiy Beskrestnov, a technology adviser to Ukraine’s president, said on Telegram that twelve of the turrets are installed in Kyiv with more to come, calling them “an innovative Ukrainian solution that shoots down (drones) using machine guns controlled by artificial intelligence.” Both sides of the war are fielding AI-powered drones — and the human factor, in Kyiv’s phrasing, is now what the machine is built to eliminate.
Gaza: the NAZA documentary’s AI allegations. The documentary NAZA — built on on-record but anonymized testimony from 24 current and former Israeli military and intelligence personnel — alleges pervasive phone monitoring, a keyword alert system, and AI systems used to identify targets with an estimated 15% error rate, plus a military shorthand called “NAZA” used to set acceptable civilian-casualty counts for planned strikes, with benchmarks cited around 20 and on one occasion up to 500, per Particle’s summary of the coverage (updated today, Oct 23).
Israel’s military and government have issued formal rebuttals, released a counter-film, and launched legal and security inquiries aimed at identifying the anonymous sources; rights and press-freedom groups say those moves risk chilling investigative journalism. And today, Al Jazeera reports an Israeli drone killed a Palestinian in Gaza; 1,439 Palestinians have been killed in Gaza since the October 2025 ceasefire began, per that report.
Quick hit — be nice to Claude: it’s policy now. Anthropic’s revamped Usage Policy, effective November 12, prohibits “sustained and needless abusive or cruel behavior” toward its models — extreme, repeated cruelty with no discernible purpose. Ordinary frustration, pushback, dark creative themes, and model testing are explicitly excluded; the stated primary enforcement is that Claude ends the conversation (a capability Anthropic added in August 2025). Anthropic told The Verge the rule covers only the extreme end — and it has not established that its models experience suffering. Announced October 8; conversation-ending is the confirmed consequence.
The Editorial
This week two governments wrote AI rules, and they wrote opposite ones. Homeland Security’s rulebook — Directive 139-08 — keeps its ban on AI mass surveillance printed on the page and hands one official the switch to turn it off, with no notice and no expiration. Italy’s Decree 179/2026, in force today, fines companies up to 7% of global turnover, voids algorithmic firings, and makes missing human oversight a crime that can mean prison. One writes the rule and keeps the exit. The other writes the rule and means it. Which one your data passes through depends on an ocean — and, in the American case, on a waiver clause.
Meanwhile the machines are already deciding. Over Kyiv, turrets with optical vision and AI “eliminate the human factor” before they fire. In Gaza, two dozen voices — on the record, unnamed — describe AI target identification with a 15% error rate and a shorthand for how many civilians a strike may cost, and the state answers with a counter-film and an inquiry into who talked. The technology keeps arriving at the same question from opposite directions: when the machine is wrong, who answers for it?
Three beats, one week: rules with an off switch, rules with teeth, and machines pulling triggers while the paperwork catches up. The off switch is held by one person. The teeth bite the companies. Nobody has named the one who answers when the gun is wrong.
— Emmett Sloane