Good morning. Anthropic rattled markets and industry leaders when the company said its new Claude AI model could pose a threat to cybersecurity systems across the world. In focus today, we look at the real risks and the “shock marketing” effect propelling the AI company into the news.
Up first
In the news
Taxes: Prime Minister Mark Carney announced a tax break on gas and diesel and said his newly secured majority will allow his government to move more quickly on economic priorities.
Accounting: Ontario’s accounting industry regulator has ruled that two partners at Deloitte LLP committed professional misconduct by failing to address allegations of fraud at Bondfield Construction Co. Ltd.
Energy: As volatile as oil futures have been, current prices suggest global shortages caused by the war in the Middle East are a fleeting problem. Sky-high prices for physical barrels, however, highlight worries over an unprecedented supply crisis.
Do electric apples keep digital doctors away?Moor Studio/iStockPhoto / Getty Images
In focus
Risk, diagnosed and treated
Weeks before disclosing a new artificial intelligence model that it says can “surpass all but the most skilled humans” at finding and exploiting software vulnerabilities, Anthropic brought together a select group of firms across Silicon Valley and Wall Street to test the “Mythos” system against critical infrastructure and assess the risks it poses to widely used software.
On April 7, Anthropic announced Project Glasswing, an initiative that gives Amazon Web Services, Apple, Google, JPMorgan Chase and other launch partners continuing access to Mythos to identify and fix vulnerabilities across critical software systems, with plans to expand participation and share findings over time. On a website dedicated to the effort – anchored by a nearly six-minute documentary featuring developers and executives from those companies – the initiative is laid out as a plan that pairs technical collaboration with subsidized access.
It also suggests a self-reinforcing business model for Anthropic: a system described as capable of exposing vulnerabilities across widely used software is also positioned as the tool needed to defend against them.
The model found “thousands of high-severity vulnerabilities,” the company said – setting off alarms across industry and government, whose leaders are now booking meetings with Anthropic executives as they race to prepare for a system that researchers found could autonomously carry out complex cyberattacks.
While there is broad agreement about the risks, the manner in which Anthropic disclosed the capabilities of Mythos – from the Ancient Greek for “utterance” or “narrative,” as the company notes – invites a closer reading of its framing.
A relief and a reckoning
Regulators and experts expressed relief that Anthropic held the model back and gave institutions advance notice before any broader release, allowing for time to test systems, identify vulnerabilities and begin adapting defences.
But the spotlight on the company in recent weeks has also revealed a new stage in the evolution of AI and its effects on society as a whole: The urgent cybersecurity risks in question strike across the institutions that underpin policy making, banking, education, stock markets, even the technology sector itself. And the solution to these risks, at least in large part, is offered by the same companies that create them – a small group of companies in Silicon Valley that decide what to release, what to share and when.
After meeting with officials from the company yesterday, Federal Artificial Intelligence Minister Evan Solomon said in a statement that Anthropic’s decision to not publicly release its powerful new AI model is the “kind of proactive approach we expect from frontier AI companies.”
Some experts say expecting the best of companies over such weighty decisions, in the absence of a clear regulatory and auditing process, is a risky proposition.
Regulators have long struggled to keep pace with rapid technological disruption. The rise of AI has been met, so far, by a fragmented set of approaches: In Europe, human rights are prioritized through the world’s first comprehensive AI Act, which bans high-risk biometric and “social-scoring” systems. In China, more targeted measures through strict state control ensure AI-generated content is labeled and remains faithful to national security.
Caught between these competing spheres, Canada is attempting to carve out a “middle path” strategy aimed at “digital sovereignty” – protecting Canadian values and data from being subsumed by U.S. tech giants.
Shock marketing
In an analysis published earlier this week, British researchers found that Mythos could autonomously exploit complex network and software vulnerabilities that would take human professionals days to complete.
Researchers who tested the model found it could autonomously complete a simulated 32-step network attack in three out of 10 attempts. But their findings came with an important bit of context: The simulations represented easier targets than real-world environments.
“This means we cannot say for sure whether Mythos Preview would be able to attack well-defended systems,” its report read.
How Anthropic disclosed the existence of Mythos – saying the model could “surpass all but the most skilled humans” at finding and exploiting software vulnerabilities, and choosing not to release it publicly – falls into the “shock marketing” approach most commonly used by distributors of horror movies, which for decades have promoted films as so overwhelming! so deeply disturbing! that audiences fainted or became ill.
So utterly affecting that you really had to see it to believe it.
Restricting access to a small group of partners, presenting Mythos as a “reckoning” for the cybersecurity industry, the company has created both fear and hype. And headlines.
Whether guided by morality, profitability, or a mixture of both, the effect is the same: high-risk decisions unfolding largely out of public view, the consequences felt across every corner of human life.
Charted
Tipping point
The Globe and Mail
Carney’s Liberals won all three of this week’s by-elections and full control of the House of Commons. Deep down, Conservative Party Leader Pierre Poilievre might welcome the new Liberal majority, Robyn Urback writes.
Quoted
It becomes less abstract, and they care about the issue more and it makes them want to learn a little bit more.
— Tracy Thomson, teacher
Through music, students get a lesson in Indigenous life and history.
Up next
More files we’re following
By the numbers: Statistics Canada reports manufacturing sales and wholesale trade data for February. Economists are expecting a modest lift after a slow start to the year. The U.S. releases the Beige Book, a colourful look at the economy on the ground level.
Before the bell: Earnings today include Bank of America Corp. and Morgan Stanley.
Morning update
Global markets were muted following yesterday’s rally, as investors evaluated a range of corporate earnings reports while monitoring the evolving situation in the Middle East.
Wall Street futures were in negative territory, while TSX futures followed sentiment lower.
Overseas, the pan-European STOXX 600 edged down 0.08 per cent. Britain’s FTSE 100 advanced 0.08 per cent, Germany’s DAX slipped 0.05 per cent and France’s CAC 40 declined 0.64 per cent.
In Asia, Japan’s Nikkei closed 0.44 per cent higher, while Hong Kong’s Hang Seng rose 0.29 per cent.
The Canadian dollar traded at 72.59 U.S. cents.